# 1stepGrow — full content > Independent tech publication from Bengaluru: technology news and analysis, honest course comparisons, career roadmaps and expert guest posts. 95 documents. Index: https://www.1stepgrow.com/llms.txt · Per-page Markdown: send `Accept: text/markdown`. --- # The 5 Ps of Data Science: Where a Busy Data Team Stalls > The 5 Ps of data science (Problem, People, Process, Platform, Portfolio): the question each forces, the failure it exposes, and why the order matters. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Sep 15, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 7 min - **Canonical URL:** https://www.1stepgrow.com/articles/5-ps-of-data-science/ ## Key takeaways - The five Ps are Problem, People, Process, Platform and Portfolio, and the order matters more than the list: a platform cannot rescue a problem nobody has defined. - Each P is worth keeping only because it names a specific failure: no decision behind the work, no owner, no repeatable path to an answer, no access to data, or no way to stop old work. - The framework describes how work gets organised, not whether the answer is right, and it cannot substitute for evaluation of the analysis itself. - A solo analyst genuinely needs three of the five; the Portfolio and most of the People discussion only start paying at a team of five or six.

The 5 Ps of data science are Problem, People, Process, Platform and Portfolio: five conditions that must hold before analytical work changes a decision. Each P names a recognisable way teams fail, and the order matters. Platform is the only P you can buy, which is why teams so often buy tooling to fix a problem nobody has defined.

This piece is for analysts, data science leads and managers whose team is busy but not landing results. For each P you get the question it forces, the symptom when it is missing and the trade-off of fixing it, then three places where the framework stops helping. Used as a diagnostic, it earns its place; filled in as a checklist, it is wallpaper.

When do the 5 Ps of data science actually help?

They help in one situation: a team is busy and nothing is landing. Somebody is always certain they know why — usually that they need better tooling, because tooling is the part you can buy. The five Ps are useful because they force you to check the cheap causes before the expensive ones.

PThe question it forcesThe symptom when it is missing
ProblemWhat decision changes because of this work?Findings that are interesting and unused
PeopleWho owns this, and who is accountable for it being right?Work restarted three times by three people
ProcessHow does a question become an answer someone trusts?Results nobody can reproduce, including their author
PlatformCan people get the data and run the work?Six weeks to first query, analysts building their own extracts
PortfolioWhich of these things should we not be doing?Fourteen half-maintained models and no capacity to start anything

Read the table in order. The first row without a concrete answer is where your problem is, and it is almost never the row the argument is about.

1. Problem: the decision, not the topic

A problem is not a topic. "Customer churn" is a topic. "Which of these 4,000 accounts should the retention team call this month, and how much better than calling them at random does that need to be to be worth the analyst time?" is a problem.

The test is whether you can complete this sentence: if the answer comes back X, we will do A; if it comes back Y, we will do B. If both branches lead to the same action, the analysis is decoration however well you execute it.

This is where data science teams waste more effort than anywhere else, and it is cheap to fix — it costs a conversation before the work starts rather than a quarter afterwards. Google's Rules of Machine Learning starts in exactly this place: "Don't be afraid to launch a product without machine learning", followed immediately by "First, design and implement metrics." Decide what you are trying to move before deciding how. What makes a good AI project applies the same judgement at project level.

The trade-off: insisting on a named decision genuinely does suppress exploratory work, and some exploratory work pays. The resolution is to timebox exploration explicitly rather than to disguise it as a project with fictional business value.

2. People: skills, but mostly ownership

The usual framing of this P is a list of roles — data scientist, data engineer, analytics engineer, ML engineer, domain expert. That list is worth having, but it is the less important half.

The important half is ownership. For any piece of work, three questions need a name attached:

Teams that cannot answer the third question produce a great deal of work that quietly stops functioning. Teams that cannot answer the second produce work nobody defends in a meeting.

Skill mix matters in one specific way: the gap between people who can produce an analysis and people who can make it run without them. That boundary decides how much of your team's capacity goes on keeping old things alive. The technical skills a data science manager actually needs sets out what it takes to hold this together.

The trade-off: hard role boundaries produce hand-off queues; no boundaries produce five people all doing the interesting part. Small teams should err towards generalists and accept the duplication.

3. Process: how a question becomes an answer twice

Process here means the repeatable path from question to trusted answer — not a governance document.

The established description of that path is CRISP-DM, whose phases are business understanding, data understanding, data preparation, modelling, evaluation and deployment. IBM's CRISP-DM guide is the most accessible primary write-up, and its framing of the first phase — gain as much insight as possible into the business goals, and meet stakeholders to establish what the work addresses — is the same point as the Problem P, arrived at from the other direction. The two frameworks stack: CRISP-DM describes one project's phases, the five Ps describe the conditions around all of them.

What makes a process real rather than documented:

That is a short list, and it is the whole of it for most teams. The failure mode is not an absence of process documents; it is documents that describe a process nobody follows.

The trade-off: process has a fixed cost per project, which is why it feels punitive on small ones. Scale it — a two-day piece of analysis needs the assumptions written down and nothing else.

4. Platform: the part you can buy, and therefore the part that gets bought first

Platform covers storage, compute, orchestration, the catalogue, and everything that decides how long it takes an analyst to get from a question to a query. Its honest measure is one number: how many days from a new joiner's first day to their first query against production data.

It gets bought first because it is purchasable, and it is the P most likely to be over-specified. Teams acquire a lakehouse, a feature store and an orchestration layer before they have enough recurring work to justify any of them, then spend a year on migration instead of analysis. Modern data engineering in 2026 covers where the genuine decisions now sit: open table formats, catalogues as the new lock-in, and object-store economics.

The other half of Platform is the part nobody demonstrates in a vendor call: what happens after something is deployed. Hidden Technical Debt in Machine Learning Systems (Sculley et al., NIPS 2015) remains the standard account of how machine learning systems accumulate maintenance cost through entanglement, undeclared consumers and configuration sprawl — costs that fall on the surrounding system, not the model. The MLOps engineer career guide describes the discipline that absorbs them, and tools and techniques for data science covers the working toolset.

The trade-off: under-investment is also real. If analysts are maintaining private extracts because the platform cannot serve them, you are paying for infrastructure anyway, just in the most expensive currency available.

5. Portfolio: deciding what not to do

Portfolio is the P most often reduced to a slide about strategic alignment, and it is the one with the sharpest practical edge: a portfolio view exists so that things can be stopped.

Every model and dashboard a team ships becomes a standing liability. It needs monitoring, it breaks when something upstream changes, and somebody has to answer questions about it. Teams that never retire anything gradually convert their capacity into maintenance, and then cannot explain why they ship nothing.

A workable portfolio practice is small:

The trade-off: this is genuinely overhead below a certain size. One or two people do not need a portfolio review; they need to remember what they built. At five or six, nobody holds it all any more, and that is when the list starts paying for itself.

What can the 5 Ps not tell you?

They cannot tell you whether an answer is right, what good looks like in your organisation, or whether a sixth condition matters more than one of the five. A framework that claims no limits is marketing.

It says nothing about whether an answer is right. Every P can be in place around a badly specified experiment or a leaking model. Correctness is a separate discipline, and the five Ps will not catch a failure in it.

It is a mnemonic, not a method. It tells you which conditions to check. It does not tell you what good looks like in your organisation, and any version that claims to is describing someone else's company.

Alliteration distorts it. Five words beginning with the same letter is a memory aid, not a finding about how work happens. Data quality, for instance, does not begin with P and matters more than at least two of these. If a sixth condition matters in your context, add it and let the mnemonic break.

Used within those limits, the five Ps do one useful job well: when a team is busy and nothing is landing, they tell you where to look first, and they stop you from buying a platform to solve a problem definition.

## Frequently asked questions ### What are the 5 Ps of data science? Problem, People, Process, Platform and Portfolio. The Problem is the decision the work should change, People are the skills and ownership around it, Process is how work moves from question to answer repeatably, Platform is the tooling and data infrastructure, and Portfolio is how competing projects are prioritised and retired. ### Is the 5 Ps framework a replacement for CRISP-DM? No, they answer different questions. CRISP-DM describes the phases of a single analysis, from business understanding through to deployment. The 5 Ps describe the conditions around the work — who does it, on what, and which projects get chosen at all. The two sit at different levels and combine cleanly. ### Which of the 5 Ps do teams get wrong most often? Problem and Portfolio. Problem fails when nobody can name the decision the analysis would change, so the result is interesting and unused. Portfolio fails when projects are started but never stopped, so a team of eight ends up maintaining fourteen half-alive models and starting nothing new. ### Does a small data team need all five Ps? A one- or two-person team needs Problem, Process and enough Platform to be repeatable. The People discussion about role boundaries and the Portfolio discussion about prioritisation start to matter at around five or six people, when nobody can hold the whole picture in their head any more. ### How do I use the 5 Ps in practice? Use it as a diagnostic rather than a plan. When a project is stuck or a team is busy without shipping, check each P in order and find the first one that does not have a concrete answer. That is almost always where the problem is, and it is rarely the one people are arguing about. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/5-ps-of-data-science/). Cite with the title, "1stepGrow" and a link._ --- # Best Data Science Courses in the USA (2026): Degrees and Certificates > The best data science courses in the USA for 2026, from accredited online master's degrees cheaper than many bootcamps to certificates for beginners. - **Author:** Ravi Pradhan — AI Expert - 11+ Yrs Experience (https://www.1stepgrow.com/authors/ravi-pradhan/) - **Published:** Sep 3, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 13 min - **Canonical URL:** https://www.1stepgrow.com/articles/best-data-science-ai-courses-usa-2026/ ## Key takeaways - Georgia Tech's OMSCS and OMSA are the best-value accredited degrees in the US, costing roughly $7,000–$13,000 in total. - UT Austin's online MSDS and MSAI offer similar value at about $10,000 plus fees, with a curriculum focused on data science or AI from the start. - Beginners should start with a low-cost Coursera certificate or the DeepLearning.AI Machine Learning Specialization before committing to a degree. - Before you enroll, confirm accreditation, the total cost including fees, and whether your employer's tuition assistance applies. The best data science courses in the USA for most working professionals in 2026 are the low-cost online master's degrees from Georgia Tech (OMSCS and OMSA) and UT Austin (MSDS and MSAI). Georgia Tech's two degrees cost roughly $7,000–$13,000 in total, so an accredited master's from a top engineering school can cost less than a three-month bootcamp. That quirk changes the decision for anyone with a STEM degree and the stamina for part-time graduate study. The University of Illinois online Master of Computer Science is the next pick for engineers who want a data science track. For a shorter credential, Stanford's AI Graduate Certificate and MIT's MicroMasters in Statistics and Data Science carry real academic weight, and beginners get the best return from the DeepLearning.AI Machine Learning Specialization or IBM's Data Science Professional Certificate. This guide is for beginners, working professionals, career changers and managers. It ranks eight options, explains how they were compared, and matches each to a type of learner. All fees are indicative ranges drawn from providers' own pages, so confirm them before applying. ## How we compared these courses This ranking is an editorial comparison of each program's published curriculum, fees, format and credential, taken from providers' official pages. I have spent more than 11 years in AI and data across EdTech, financial services and technology. We read each program with one question in mind: what does a working professional in the US actually get for their time and money? We did not survey students or interview alumni, and the ratings are our editorial scores, not aggregated reviews. We weighed six factors: - **Curriculum depth.** Coverage of statistics, machine learning, deep learning and data engineering, and whether the program has caught up with generative AI. - **Format.** Asynchronous or live, and whether the pace is realistic alongside a full-time job. - **Accreditation and credential recognition.** An accredited degree, a graduate certificate with academic credit, and a non-credit professional certificate are very different things to an employer. - **Total cost.** The full program price including fees, not the per-credit headline. - **Support.** Teaching assistants, forums, office hours and any career services. - **Fit for working professionals.** Flexibility to pause, pay per course or study across several years. Degrees rank above certificates here because they score higher on credential recognition and depth, not because certificates are poor value. A $49-a-month certificate that you finish can be worth more to you than a degree you drop after one semester. Read the ratings as a guide to what each program does well, then use the persona section below to match one to your situation. ### 1. Georgia Tech Online Master's (OMSCS and OMSA) Accredited, rigorous graduate degrees in computer science or analytics at a fraction of typical master's tuition. - **Provider:** Georgia Institute of Technology - **Price:** Approx. $7,000–$8,000 total (OMSCS); approx. $12,000–$13,000 total (OMSA) - **Duration:** Typically 2–3 years part-time (up to 6 years allowed) - **Level:** Advanced - **1stepGrow score:** 4.8/5 - **Best for:** Working professionals with a STEM background who want a full master's degree without taking on debt - **Website:** https://omscs.gatech.edu/ **Pros** - Same degree as on-campus students, from a top-ranked engineering school - OMSCS offers a machine learning specialization; OMSA has a Computational Data Analytics track - Asynchronous delivery fits around a full-time job in any US time zone **Cons** - Demanding workload; many students take one course per semester - Large classes mean support comes mostly through TAs and forums rather than faculty **Verdict:** The best-value data science or AI credential in the country for anyone who can handle graduate-level math and programming. Choose OMSA for analytics and business-facing roles, OMSCS for machine learning engineering. ### 2. Online MS in Data Science / MS in Artificial Intelligence Focused online master's degrees in data science or AI from a top public research university. - **Provider:** The University of Texas at Austin - **Price:** Approx. $10,000 plus fees total (international fees may apply) - **Duration:** 10 courses (30 credit hours), studied part-time - **Level:** Advanced - **1stepGrow score:** 4.7/5 - **Best for:** Professionals who want a dedicated data science or AI degree rather than a general computer science one - **Website:** https://cdso.utexas.edu/msds **Pros** - MSDS is run jointly by the Statistics and Data Sciences and Computer Science departments - MSAI electives cover deep learning, reinforcement learning and natural language processing - Asynchronous, instructor-paced courses built for working adults **Cons** - Assumes strong math and programming preparation - The MSAI is newer and has a shorter track record than long-running programs **Verdict:** Georgia Tech's closest rival on value, with a curriculum focused on data science or AI from day one. ### 3. Online Master of Computer Science (Data Science track) A coursework-only CS master's with a data science track, delivered on Coursera and assessed by Illinois faculty. - **Provider:** University of Illinois Urbana-Champaign (on Coursera) - **Price:** Approx. $20,000–$25,000 total, depending on residency - **Duration:** 8 courses (32 credit hours); 1–5 years - **Level:** Advanced - **1stepGrow score:** 4.5/5 - **Best for:** Software engineers moving toward machine learning, data mining and cloud data work - **Website:** https://siebelschool.illinois.edu/academics/graduate/professional-mcs/online-master-computer-science **Pros** - The data science track covers machine learning, data mining, data visualization and cloud computing - Pay-as-you-go tuition by course, and you can take a term off - No thesis, which suits people working full time **Cons** - Roughly two to three times the total cost of Georgia Tech or UT Austin - Platform-based delivery can feel less personal than live cohort formats **Verdict:** A strong degree for engineers who value the Illinois name and the flexibility of online study, if the higher price is acceptable. ### 4. Artificial Intelligence Graduate Certificate Four graduate-level Stanford AI courses, taken online for credit. - **Provider:** Stanford Online - **Price:** Approx. $20,000–$25,000+ total (charged per unit) - **Duration:** 4 courses, completed within 3 years - **Level:** Advanced - **1stepGrow score:** 4.5/5 - **Best for:** Engineers and technical leads who want deep AI coursework and a Stanford transcript without a full degree - **Website:** https://online.stanford.edu/programs/artificial-intelligence-graduate-certificate **Pros** - Graduate courses in areas such as machine learning, natural language processing and computer vision - Earns academic credit and a Stanford graduate certificate - Shorter commitment than a master's degree **Cons** - Expensive per course compared with online master's degrees - Demands strong math, probability and programming skills **Verdict:** The premium choice for experienced technical professionals whose employer will pay. For self-funded learners, Georgia Tech or UT Austin delivers a full degree for less. ### 5. MicroMasters Program in Statistics and Data Science MIT-level probability, statistics and machine learning online, at a very low total price. - **Provider:** MIT (MITx) - **Price:** Approx. $1,350–$1,500 for the full credential - **Duration:** 4 courses plus a proctored capstone exam; typically a year or more - **Level:** Advanced - **1stepGrow score:** 4.6/5 - **Best for:** Self-directed learners who want rigorous statistical foundations and an MIT credential at low cost - **Website:** https://micromasters.mit.edu/ds/ **Pros** - Probability and machine learning taught at the pace and rigor of MIT's on-campus courses - Very low cost for a credential from MIT - Some universities accept the credential toward a master's degree **Cons** - Math-heavy and easy to abandon without external accountability - A MicroMasters is not a degree, so its value depends on employers and pathway schools recognizing it **Verdict:** The best way to test whether you can handle graduate-level data science before paying for a degree, and a credible credential in its own right. ### 6. Online MS in Artificial Intelligence / MS in Data Science Part-time online master's degrees from Johns Hopkins, built for engineers working full time. - **Provider:** Johns Hopkins Engineering for Professionals - **Price:** Approx. $5,400–$5,500 per course (about $55,000 for 10 courses) - **Duration:** 10 courses; up to 5 years - **Level:** Advanced - **1stepGrow score:** 4.2/5 - **Best for:** Professionals whose employer tuition assistance will cover most of the cost - **Website:** https://ep.jhu.edu/programs/artificial-intelligence/ **Pros** - Designed specifically for part-time working engineers - Strong name recognition, particularly in health and government-related sectors - Choice of AI or data science degree paths **Cons** - Among the most expensive options on this list - Value depends heavily on employer funding **Verdict:** A solid, well-recognized degree if someone else is paying. Self-funded learners get a comparable credential for far less at Georgia Tech or UT Austin. ### 7. Machine Learning Specialization Andrew Ng's three-course introduction to modern machine learning, in Python. - **Provider:** DeepLearning.AI and Stanford Online (on Coursera) - **Price:** Approx. $49–$59/month on Coursera (financial aid available) - **Duration:** 3 courses; about 2 months at 10 hours a week - **Level:** Beginner - **1stepGrow score:** 4.4/5 - **Best for:** Beginners and analysts who want a clear, well-taught first course in machine learning - **Website:** https://www.coursera.org/specializations/machine-learning-introduction **Pros** - Covers supervised learning, neural networks, decision trees, unsupervised learning, recommender systems and reinforcement learning basics - Hands-on labs with NumPy, scikit-learn and TensorFlow - A natural lead-in to DeepLearning.AI's Deep Learning Specialization and generative AI courses **Cons** - The certificate carries limited weight with employers on its own - Light on SQL, data engineering and statistics for analysis **Verdict:** The best low-cost entry point into machine learning. Pair it with a portfolio project to make it count. ### 8. IBM Data Science Professional Certificate A broad, beginner-friendly tour of the data science workflow, from Python and SQL to basic machine learning. - **Provider:** IBM (on Coursera) - **Price:** Approx. $49/month on Coursera - **Duration:** Several months part-time, self-paced - **Level:** Beginner - **1stepGrow score:** 4/5 - **Best for:** Complete beginners testing whether data science suits them before a bigger commitment - **Website:** https://www.coursera.org/professional-certificates/ibm-data-science **Pros** - No prerequisites and a gentle on-ramp - Covers Python, SQL, data analysis, visualization and introductory machine learning - Low monthly cost, so finishing faster costs less **Cons** - Introduces topics rather than teaching them in depth - Widely held, so it does little to make a resume stand out **Verdict:** A sensible, low-risk first step. If analytics roles appeal more, Google's Data Analytics Professional Certificate is the equivalent starting point. ## How do the eight US options compare on cost? Totals range from about $1,350 for the MITx MicroMasters to roughly $55,000 for Johns Hopkins, with the Georgia Tech and UT Austin degrees near the low end and the Coursera certificates billed monthly. | Course | Format | Duration | Indicative fee | Best for | |---|---|---|---|---| | Georgia Tech OMSCS / OMSA | Online, asynchronous degree | 2–3 years part-time | ~$7,000–$13,000 total | STEM professionals wanting a full degree | | UT Austin MSDS / MSAI | Online, asynchronous degree | 10 courses, part-time | ~$10,000 + fees | A focused data science or AI degree | | UIUC Online MCS (Data Science track) | Online degree on Coursera | 1–5 years | ~$20,000–$25,000 | Software engineers moving into ML | | Stanford AI Graduate Certificate | Online graduate courses | 4 courses within 3 years | ~$20,000–$25,000+ | Experienced engineers with employer funding | | MITx MicroMasters in Statistics and Data Science | Online, scheduled courses | 4 courses + capstone exam | ~$1,350–$1,500 | Rigorous, low-cost foundations | | Johns Hopkins EP MS in AI / Data Science | Online, part-time degree | 10 courses, up to 5 years | ~$55,000 total | Employer-sponsored professionals | | DeepLearning.AI Machine Learning Specialization | Self-paced online | ~2 months | ~$49–$59/month | Beginners starting machine learning | | IBM Data Science Professional Certificate | Self-paced online | Several months | ~$49/month | Complete beginners | As of September 2026, prices are indicative. They change regularly and can differ for residents, non-residents and international students, so confirm the current figure on each provider's page before you apply. ## How do you decide between the best data science courses in the USA? ### Check accreditation first For any degree or graduate certificate, confirm that the school holds institutional accreditation from an agency [recognized by the US Department of Education](https://www.ed.gov/laws-and-policy/higher-education-laws-and-policy/college-accreditation) or CHEA. All the universities on this list do. Accreditation affects whether federal student aid applies, whether your employer will reimburse you and whether credits transfer later. Non-credit certificates from Coursera providers are not accredited degrees. They can still be useful, but they are a different category of credential. ### Work out the true total cost Per-credit pricing makes programs look cheaper than they are. Add tuition, per-term technology fees and any international or out-of-state surcharges across the whole program. Also consider how long you will realistically take. A pay-per-course model such as Illinois's lets you pause without penalty, which lowers the risk if your job gets busy. ### Use employer and government funding Many US employers offer tuition assistance, and up to $5,250 a year of employer-provided educational assistance [can be tax-free for the employee](https://www.irs.gov/publications/p15b). Read the policy for approved program types, minimum grades and repayment clauses if you leave. Veterans should check programs with the VA's [GI Bill Comparison Tool](https://www.va.gov/education/gi-bill-comparison-tool/). Federal student aid is generally available for eligible degree programs at accredited schools, but not for most non-credit certificates or bootcamps. ### Consider time zones and pace Georgia Tech, UT Austin and Illinois are largely asynchronous, which suits shift workers and people on the West Coast or overseas. Programs with live elements usually schedule them on Eastern or Pacific time. Be honest about your weekly hours. Many working students take one graduate course per term, and taking two alongside a demanding job is where many people struggle. ### Match the course to the role you want The right program depends on the job you are aiming for: - **Data analyst:** a professional certificate plus strong SQL and a portfolio is often enough to get interviews. Georgia Tech's OMSA is a strong later step. - **Data scientist:** you need statistics and machine learning in depth. UT Austin's MSDS, OMSA's Computational Data Analytics track or the MITx MicroMasters fit best. - **Machine learning or AI engineer:** you need software engineering alongside ML. OMSCS with the machine learning specialization, UT Austin's MSAI, the Illinois MCS and Stanford's AI Graduate Certificate are the closest fits. - **Data engineer:** none of the programs above is a pure data engineering degree. Pair a computer science program's database and cloud courses with hands-on work in Spark, dbt and a cloud data warehouse. ### Questions to ask before you enroll - What is the total cost, including all fees, at the pace I can realistically manage? - Is the school institutionally accredited, and will my employer's tuition assistance policy accept this program? - What math and programming background do successful students have, and is there a readiness course or self-assessment? - How much of the teaching is live, and in which time zone? - Can I pause for a term without losing my place or paying a penalty? - For bootcamps, what share of everyone who enrolled, not just graduates, found a relevant job within six months? ### Know where bootcamps fit Bootcamps such as Flatiron School and General Assembly run intensive data science programs over roughly three months full time, or longer part time. List tuition commonly sits in the mid-teens of thousands of dollars. They suit people who need structure, a cohort and career coaching more than a formal credential. Ask for outcomes data that counts everyone who enrolled, not just graduates who job-searched. Read financing and income-share terms line by line. For a fuller look at the economics, see [is a data science bootcamp worth it](https://www.1stepgrow.com/articles/is-a-data-science-bootcamp-worth-it). ### Look at other well-known options Harvard offers the HarvardX Professional Certificate in Data Science on edX, a low-cost, R-based series. Harvard Extension School also offers graduate certificates. Community colleges are an underrated starting point for fundamentals, with credits that can transfer to a bachelor's program. Neither made our top eight, but both are worth a look for specific needs. ### A note for international students Most online programs accept international applicants, sometimes at different fees. An online degree studied from abroad does not create US work authorization. F-1 students completing eligible STEM degrees in the US may qualify for OPT and the STEM OPT extension. Rules change, so confirm with the school's international office and official government sources. ## Which course should you choose? ### If you are a complete beginner Start small and cheap. Take the IBM Data Science Professional Certificate or Google's Data Analytics Professional Certificate to find out whether you enjoy the work. Then take the DeepLearning.AI Machine Learning Specialization. Build two or three portfolio projects on real, messy data. If you want to test whether you can handle graduate-level statistics, try one MITx MicroMasters course before applying to a degree. ### If you are a working professional in a technical role Apply to Georgia Tech or UT Austin. Choose OMSA if you want analytics and decision-science roles. Choose OMSCS or UT Austin's MSAI for machine learning engineering, and UT Austin's MSDS for a balanced statistics-and-computing degree. If you already work as a software engineer and prefer the Illinois brand or the Coursera platform, the UIUC online MCS with its data science track is worth the higher price. Most of these programs do not lean heavily on standardized tests; OMSA, for example, does not require the GRE or GMAT. You will need to show evidence of math and programming preparation, such as prior coursework or a completed online course. ### If you are switching careers Your fastest route is usually through a data analyst role, not straight to data scientist. Build SQL, spreadsheet and dashboard skills with a professional certificate, then add Python and statistics. A bootcamp can help if you need external structure, but check its outcomes data carefully. Once you have a data-adjacent job, an online master's funded by your employer is a strong next step. ### If you are a manager or team lead You probably need fluency rather than engineering depth. A short course such as the DeepLearning.AI Machine Learning Specialization gives you the vocabulary to work well with technical teams. If you lead technical staff and your employer will fund it, Stanford's AI Graduate Certificate provides serious depth and a credential that signals it. ## Related reading For the wider picture of studying and working in data in the US, including tech hubs and funding routes, see our [USA tech courses and careers hub](https://www.1stepgrow.com/regions/usa/). If you are still deciding on a direction, our [data scientist roadmap](https://www.1stepgrow.com/articles/data-scientist-roadmap-2026) sets out the skills in the order to learn them. For AI-specific options beyond the US, see the [best AI courses in 2026](https://www.1stepgrow.com/articles/best-ai-courses-2026). If you are weighing a bootcamp, [is a data science bootcamp worth it](https://www.1stepgrow.com/articles/is-a-data-science-bootcamp-worth-it) goes deeper on the economics. ## Frequently asked questions ### What is the best data science course in the USA for working professionals? For most working professionals with a STEM background, Georgia Tech's OMSA or OMSCS and UT Austin's online MSDS or MSAI offer the best mix of rigor, recognition and price. Both are accredited, asynchronous master's degrees that cost roughly $7,000–$13,000 in total. If you only want a few courses, Stanford's AI Graduate Certificate or MIT's MicroMasters in Statistics and Data Science are strong alternatives. ### Is a data science bootcamp or an online master's better in the US? For people who can manage graduate-level math, an online master's from Georgia Tech or UT Austin often costs less than a bootcamp and carries more weight with employers. Bootcamps suit career switchers who need an intensive, structured push over a few months. Many bootcamps list tuition in the mid-teens of thousands of dollars, so ask for full outcomes data before paying. ### Can beginners take an online data science master's in the USA? Most online master's programs expect a bachelor's degree plus solid preparation in math and programming, so complete beginners usually need groundwork first. A low-cost certificate such as the IBM Data Science Professional Certificate or the DeepLearning.AI Machine Learning Specialization is a sensible way to build that base. MIT's MicroMasters is a good test of graduate-level readiness. ### How much do data science courses cost in the USA? Costs range from about $49 a month for Coursera professional certificates to roughly $7,000–$13,000 for Georgia Tech's online degrees and $50,000 or more for some part-time master's programs. Graduate certificates such as Stanford's typically cost $20,000 or more. The prices here are indicative, so confirm current fees on each provider's page. ### Will my employer pay for a data science course? Many US employers offer tuition assistance, and federal tax rules allow up to $5,250 a year of employer-provided educational assistance to be tax-free for the employee. Policies differ on which programs qualify and whether you must repay if you leave. Accredited degrees are usually the easiest to get approved. ### Are AI courses in the USA open to international students? Most online programs accept international applicants, although some charge different fees. Studying online from abroad does not give you US work authorization. F-1 students who complete eligible STEM degrees in the US may qualify for OPT and the STEM OPT extension, so check the rules with the school's international office. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/best-data-science-ai-courses-usa-2026/). Cite with the title, "1stepGrow" and a link._ --- # Best Data Science Courses in the UK for 2026, Plus Fee-Free Routes > The best data science courses in the UK for 2026: online AI and data science MSc programmes, apprenticeships and free Skills Bootcamps compared on cost and fit. - **Author:** Ravi Pradhan — AI Expert - 11+ Yrs Experience (https://www.1stepgrow.com/authors/ravi-pradhan/) - **Published:** Sep 2, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 13 min - **Canonical URL:** https://www.1stepgrow.com/articles/best-data-science-ai-courses-uk-2026/ ## Key takeaways - York's and Bath's online AI master's degrees are the best-value university options in the UK, at roughly £11,000–£13,000 in total. - Imperial's online MSc Machine Learning and Data Science is the most rigorous option, but it costs about £39,900 and needs a strong quantitative degree. - Multiverse apprenticeships and government-funded Skills Bootcamps let eligible learners train without paying tuition. - Check degree-awarding status, total cost and funding eligibility before enrolling, and verify any visa questions on GOV.UK. The best data science courses in the UK for most graduates in 2026 are the online master's degrees from the University of York (MSc Computer Science with Artificial Intelligence) and the University of Bath (MSc Artificial Intelligence). Both are fully online and part-time, and each costs roughly £11,000–£13,000 in total, about a third of Imperial's £39,900 online MSc. Imperial College London remains the most rigorous option, and Edinburgh's online data science MSc programmes offer flexible exit points. The bigger UK advantage, though, is funding that most countries lack: if you are employed, a Multiverse apprenticeship or a government-funded Skills Bootcamp can train you without tuition fees. This guide is for graduates, career changers, employees and managers deciding between a degree, an apprenticeship and a bootcamp. It ranks eight options, explains how they were compared, and matches each to a type of learner. All fees are indicative ranges drawn from providers' own pages, so confirm them before applying. ## How we compared these courses This ranking is an editorial comparison of each programme's published curriculum, fees, format and credential, taken from providers' official pages. I have spent more than 11 years in AI and data across EdTech, financial services and technology. We read each programme with one question in mind: what does a UK learner actually get for their time and money? We did not survey students or interview alumni, and the ratings are our editorial scores, not aggregated reviews. We weighed six factors: - **Curriculum depth.** Coverage of statistics, machine learning, deep learning and data engineering, and whether the programme has caught up with generative AI. - **Format.** Asynchronous or live, and whether the pace is realistic alongside a full-time job. - **Credential recognition.** A UK degree from a recognised awarding body, an apprenticeship standard, and a certificate of attendance mean very different things to an employer. - **Total cost.** The full price including every module, not the per-module headline. - **Support.** Tutors, forums, live sessions and any careers or employer links. - **Fit for working professionals.** Flexible start dates, pay-per-module fees and the option to stop at an intermediate award. Degrees rank above apprenticeships and bootcamps here because they score higher on credential recognition and depth, not because the other routes are poor value. A free Skills Bootcamp you finish, or an apprenticeship your employer pays for, can be worth more to you than an MSc you abandon after one module. Read the ratings as a guide to what each programme does well, then use the persona section below to match one to your situation. ### 1. Online MSc Computer Science with Artificial Intelligence A 100% online conversion MSc that takes graduates from other disciplines into computer science and AI. - **Provider:** University of York - **Price:** Approx. £11,000 total (paid per module) - **Duration:** 2 years part-time - **Level:** Intermediate - **1stepGrow score:** 4.6/5 - **Best for:** Graduates from non-computing backgrounds who want a recognised master's in AI while working - **Website:** https://online.york.ac.uk/msc-computer-science-with-artificial-intelligence/ **Pros** - Explicitly designed for people without a computing degree; entry is a 2:2 or equivalent - Covers programming, algorithms, data analytics, machine learning and applied AI - Pay-per-module fees spread the cost over the course **Cons** - As a conversion degree it trades some advanced ML depth for breadth - Two years of self-directed online study needs steady discipline **Verdict:** The most accessible route to a Russell Group master's in AI for career changers, at a fair price. ### 2. Online MSc Artificial Intelligence A specialist online AI master's covering machine learning, robotics and how humans work with intelligent machines. - **Provider:** University of Bath - **Price:** Approx. £13,000 total (paid per unit) - **Duration:** Minimum 2 years 3 months; January, May and September starts - **Level:** Intermediate - **1stepGrow score:** 4.6/5 - **Best for:** Quantitatively strong graduates who want a focused AI degree rather than a broad conversion course - **Website:** https://online.bath.ac.uk/online-courses/msc-artificial-intelligence **Pros** - Curriculum spans programming, machine learning, robotics and human–AI interaction, with deep learning electives - Three intakes a year - Pay-as-you-go unit fees; eligible home students may be able to use a Postgraduate Master's Loan **Cons** - Requires demonstrable maths (algebra and calculus) if your degree was not quantitative - Fees may rise year on year during the course **Verdict:** A well-structured, specialist AI MSc at a similar price to York. Choose Bath if you already have the maths and want AI depth from the start. ### 3. MSc Machine Learning and Data Science (Online) A mathematically rigorous, part-time online MSc from one of the UK's leading science universities. - **Provider:** Imperial College London - **Price:** Approx. £39,900 total (2026 entry; same for home and overseas students) - **Duration:** 2 years part-time - **Level:** Advanced - **1stepGrow score:** 4.5/5 - **Best for:** STEM graduates who want the most technically demanding online machine learning degree in the UK - **Website:** https://www.imperial.ac.uk/study/courses/postgraduate-taught/machine-learning-data-science/ **Pros** - Covers supervised and unsupervised learning, deep learning, Bayesian methods and big data with PySpark - The Imperial name carries weight with UK and international employers - Substantial research project in the final term **Cons** - Roughly three times the price of York or Bath - Needs a 2:1 in a quantitative subject, and applications for 2026 entry have closed, so check the next intake **Verdict:** The top pick for rigour if you have a strong maths background and the budget. Most other learners get better value from York or Bath. ### 4. Data Science Technologies / Applied Data Science (Online Learning) MSc Fully online data science master's programmes from Edinburgh, with PgCert and PgDip exit points. - **Provider:** University of Edinburgh - **Price:** Varies by award (PgCert, PgDip or MSc); check the programme page - **Duration:** 1 year full-time or 3 years part-time - **Level:** Intermediate - **1stepGrow score:** 4.4/5 - **Best for:** Professionals who want Edinburgh's data science reputation, with the option to stop at a PgCert or PgDip - **Website:** https://study.ed.ac.uk/programmes/postgraduate-taught/1136-data-science-technologies-online-learning **Pros** - Combines statistical analysis, machine learning, programming and data visualisation - Flexible awards let you study to PgCert, PgDip or a full MSc - Applied Data Science (first intake September 2026) is aimed at learners without prior programming experience **Cons** - Data Science Technologies expects a numerate degree or relevant data or programming experience - The long-running Data Science, Technology and Innovation MSc closed to new applicants after 2025, so older reviews may describe a different programme **Verdict:** A strong, flexible choice for learners who want to control how far they go. Pick Applied Data Science if you are new to programming. ### 5. BSc Computer Science (Machine Learning and AI or Data Science specialism) A full online bachelor's degree with specialisms in machine learning and AI, or data science. - **Provider:** University of London (academic direction from Goldsmiths), on Coursera - **Price:** Approx. £13,000–£21,000 total, depending on where you live - **Duration:** 3–6 years - **Level:** Beginner - **1stepGrow score:** 4.3/5 - **Best for:** Learners without a degree who want a recognised UK bachelor's in computing while working - **Website:** https://www.london.ac.uk/study/courses/undergraduate/bsc-computer-science **Pros** - Modules designed by the Goldsmiths academics who teach the on-campus programme - Specialist awards in machine learning and AI, or data science - Flexible pace, with fees that vary by region **Cons** - A multi-year commitment, and too slow for anyone who already has a degree - Online study with tutor support rather than small-group teaching **Verdict:** The best UK option for someone starting without a degree. Graduates who already hold one should look at a conversion MSc instead. ### 6. AI and Machine Learning Fellowship (Level 6 apprenticeship) Learn machine learning engineering on the job, funded through your employer's apprenticeship levy. - **Provider:** Multiverse - **Price:** No tuition fee for the apprentice; funded by the employer through the apprenticeship levy - **Duration:** Approx. 16 months of learning, plus assessment - **Level:** Intermediate - **1stepGrow score:** 4.3/5 - **Best for:** Employees who want to move into data or AI roles without leaving their job or paying fees - **Website:** https://www.multiverse.io/en-GB/employers/courses/ai-and-machine-learning-fellowship **Pros** - No tuition cost to you, and you keep earning a salary - A blend of live workshops, self-paced content and applied projects in your own role - Lower-level data programmes, such as the Level 4 Data Fellowship, suit people starting out **Cons** - Only available if your employer agrees to fund and support it - Eligibility rules apply, including UK right to work and residency requirements **Verdict:** If your employer pays the apprenticeship levy, this is one of the most cost-effective ways to build credible ML skills in the UK. Ask your HR or L&D team first. ### 7. Skills Bootcamps in data and digital skills Free, government-funded short courses in data and digital skills, with a job interview on completion. - **Provider:** Department for Education (England), delivered by approved training providers - **Price:** Free to eligible individuals; employers contribute when training existing staff - **Duration:** Up to 16 weeks - **Level:** Beginner - **1stepGrow score:** 4.1/5 - **Best for:** Career switchers in England who need a fast, fee-free route into data analysis or data engineering - **Website:** https://find-employer-schemes.education.gov.uk/schemes/skills-bootcamps **Pros** - No tuition fee for eligible learners aged 19 and over living in England - A job interview is offered on completion - Online, classroom and blended formats are available **Cons** - A short course cannot replace a degree for most data scientist roles - Availability and quality vary by region and provider, and eligibility is limited to England **Verdict:** The best free starting point in England for analyst and junior data engineering roles. Check the provider's syllabus and employer links before applying. ### 8. Oxford Artificial Intelligence Programme A six-week online programme on AI strategy and implementation, designed for business leaders. - **Provider:** Saïd Business School, University of Oxford - **Price:** Approx. £2,000–£2,500 - **Duration:** 6 weeks online - **Level:** All levels - **1stepGrow score:** 4/5 - **Best for:** Managers and executives who need to lead AI initiatives rather than build models - **Website:** https://www.sbs.ox.ac.uk/programmes/online-learning/oxford-artificial-intelligence-programme **Pros** - Designed for business audiences rather than programmers - Weekly assignments with no formal exam - Short enough to complete alongside a senior role **Cons** - Leads to a certificate of attendance, not an academic qualification - Will not prepare you for hands-on technical roles **Verdict:** A credible, time-efficient AI primer for leaders. Technical staff should choose one of the MSc programmes instead. ## How do the eight UK options compare on cost? The online MSc degrees run from about £11,000 at York to £39,900 at Imperial, the apprenticeship and Skills Bootcamps cost eligible learners nothing in tuition, and the Oxford programme is a short executive course rather than a qualification. | Course | Format | Duration | Indicative fee | Best for | |---|---|---|---|---| | York online MSc Computer Science with AI | Online, part-time degree | 2 years | ~£11,000 total | Career changers from other disciplines | | Bath online MSc Artificial Intelligence | Online, part-time degree | 2 years 3 months minimum | ~£13,000 total | Quantitative graduates wanting AI depth | | Imperial online MSc Machine Learning and Data Science | Online, part-time degree | 2 years | ~£39,900 total | STEM graduates wanting maximum rigour | | Edinburgh Data Science Technologies / Applied Data Science MSc | Online, full- or part-time | 1–3 years | Varies by award | Flexible PgCert, PgDip or MSc study | | University of London BSc Computer Science | Online degree on Coursera | 3–6 years | ~£13,000–£21,000 total | Learners without a first degree | | Multiverse AI and Machine Learning Fellowship | Apprenticeship, in work | ~16 months + assessment | No fee to apprentice | Employees with employer backing | | Skills Bootcamps (England) | Online, classroom or blended | Up to 16 weeks | Free if eligible | Fast, fee-free career switch | | Oxford Artificial Intelligence Programme | Online executive programme | 6 weeks | ~£2,000–£2,500 | Managers leading AI initiatives | As of September 2026, prices are indicative. They change regularly and can differ by fee status and location, so confirm the current figure on each provider's page before you apply. ## How do you narrow down the best data science courses in the UK? ### Check who awards the qualification For any degree, confirm the provider is a recognised UK degree-awarding body. [GOV.UK explains how to check](https://www.gov.uk/check-university-award-degree). All the universities on this list qualify. For bootcamps and short courses, ignore the branding and look at the syllabus, who teaches it and which employers are involved. ### Work out the full cost and your funding options Most online MSc programmes charge per module, which spreads the cost but makes the total easy to underestimate. Add every module, the dissertation and any annual fee increases. Eligible home students may be able to use a [Postgraduate Master's Loan](https://www.gov.uk/masters-loan/eligibility), including for distance-learning courses if they live in England when the course starts, so check the Student Finance rules for your situation. Some universities also offer alumni discounts. ### Ask your employer about apprenticeships If you are employed, ask your HR or learning and development team whether your employer uses its apprenticeship levy or funding for data and AI apprenticeships. Providers such as Multiverse run programmes at several levels. Degree apprenticeships from universities can lead to a full bachelor's or master's without tuition fees for the apprentice. Apprenticeship funding rules have been reformed recently, so confirm current eligibility before you plan around one. ### Use Skills Bootcamps if you live in England Skills Bootcamps are free for eligible adults aged 19 and over who live in England, last up to 16 weeks and include a job interview on completion. Data analysis and data engineering courses are among the options. Quality depends on the provider, so compare syllabuses and ask which employers take part in the interviews. Scotland, Wales and Northern Ireland run their own skills schemes, so check what is available locally. ### Match the course to the role you want - **Data analyst:** a Skills Bootcamp or a Level 4 data apprenticeship, plus strong SQL, Power BI or Tableau and a portfolio, is often enough to get interviews. - **Data scientist:** you need statistics and machine learning in depth. Bath, Imperial or Edinburgh's Data Science Technologies MSc fit best, with York as the entry point if you are coming from another subject. - **Machine learning or AI engineer:** you need software engineering alongside ML. York's computer science grounding, Bath's specialist AI curriculum or Multiverse's Level 6 AI and Machine Learning Fellowship are the closest fits. - **Data engineer:** Northcoders and Makers focus directly on data engineering, and some Skills Bootcamps cover it. Pair any of them with hands-on cloud work. ### Questions to ask before you enrol - What is the total cost, including every module and the dissertation, at the pace I can realistically manage? - Is the provider a recognised degree-awarding body, and could my employer fund this through an apprenticeship instead? - What maths and programming background do successful students have, and is there a preparatory course? - How much teaching is live, and at what times in UK time? - Can I pause between modules, or leave with a PgCert or PgDip if my circumstances change? - For bootcamps, what share of everyone who enrolled, not just graduates, found a relevant job within six months? ### Compare private bootcamps carefully Le Wagon London runs a Data Science & AI bootcamp in full-time and part-time formats. Makers offers a data engineering bootcamp, and Manchester-founded Northcoders runs a 13-week Data Engineering, AI & Machine Learning bootcamp. These suit people who want an intensive, cohort-based push. Check the total cost, any income-share terms and outcome figures that count everyone who enrolled. ### Consider part-time degrees and continuing education The Open University offers a part-time BSc (Hons) Data Science for people who want a full degree studied flexibly over several years. The University of Cambridge's Institute of Continuing Education runs part-time courses for adult learners. Neither made our top eight, but both suit specific needs. ### Plan around time zones and your workload Most online MSc programmes are asynchronous, with any live sessions on UK time (GMT or BST). That suits UK learners and much of Europe but can be awkward from Asia or the Americas. Budget your weekly hours honestly. One module at a time alongside a full-time job is realistic for most people. ### A note for international students The [Graduate Route](https://www.gov.uk/graduate-visa) applies to eligible degrees completed while studying in the UK on a Student visa, not online study from abroad, and applications made from 1 January 2027 get 18 months rather than two years. Student visas do not cover distance learning either. Skilled Worker visa requirements change regularly. Check current rules on GOV.UK rather than relying on course marketing. ## Which course should you choose? ### If you are a complete beginner If you do not have a degree, the University of London's online BSc Computer Science with a machine learning and AI or data science specialism gives you a recognised qualification at your own pace. If you live in England and want a faster start, a free Skills Bootcamp in data analysis is the lowest-risk way to find out whether you enjoy the work. Then build a small portfolio of projects on real data. ### If you are a working professional with a degree Apply to York if your degree is in another subject, or to Bath if you already have strong maths and want AI depth from the start. If you have a first-class or strong 2:1 degree in a quantitative subject and the budget, Imperial offers the most rigorous online programme in the UK. Edinburgh suits people who want the option to stop at a PgCert or PgDip. Bath has January, May and September starts, so you rarely wait long for an intake. Imperial's applications for 2026 entry have already closed, so plan ahead if it is your first choice. ### If you are switching careers The most reliable route into data in the UK is usually through a data analyst role. If you are employed, ask whether your employer will fund a Multiverse apprenticeship, since you keep your salary and pay no tuition. If you are between roles in England, a Skills Bootcamp offers free training and an interview. York's conversion MSc is a strong next step once you are sure data is the right direction. ### If you are a manager or team lead You probably need fluency rather than engineering depth. The Oxford Artificial Intelligence Programme covers AI strategy and implementation in six weeks and is designed for business audiences. If you manage technical staff and want deeper understanding, a single module or PgCert from Edinburgh's online programmes is a more substantial option. ## Related reading For the wider picture of studying and working in data in the UK, including tech hubs and funding routes, see our [UK tech courses and careers hub](https://www.1stepgrow.com/regions/uk/). If you are still choosing a direction, [data analyst vs data scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist) explains the day-to-day differences. For AI-specific options beyond the UK, see the [best AI courses in 2026](https://www.1stepgrow.com/articles/best-ai-courses-2026). If you are weighing a bootcamp, [is a data science bootcamp worth it](https://www.1stepgrow.com/articles/is-a-data-science-bootcamp-worth-it) goes deeper on the economics. ## Frequently asked questions ### What is the best data science course in the UK for working professionals? For graduates who are working, the University of York's online MSc Computer Science with Artificial Intelligence and the University of Bath's online MSc Artificial Intelligence offer the best balance of recognition, flexibility and cost, at roughly £11,000–£13,000 in total. If your employer pays the apprenticeship levy, a Multiverse data or AI apprenticeship avoids tuition fees altogether. ### Can I study data science in the UK for free? Yes, in some cases. Adults aged 19 or over who live in England and meet the eligibility rules can take free Skills Bootcamps of up to 16 weeks in data and digital skills, with a job interview on completion. Apprenticeships are also free to the apprentice, because the employer funds the training. ### Do I need a computer science degree for an AI master's in the UK? Not always. York's online MSc Computer Science with Artificial Intelligence is designed for graduates from other disciplines, and Edinburgh's Applied Data Science (Online Learning) MSc is aimed at learners without programming experience. Bath's programme accepts non-computing graduates who can show quantitative skills, while Imperial's expects a 2:1 in a subject such as maths, physics or engineering. ### How much does an online AI or data science MSc cost in the UK? York and Bath cost roughly £11,000–£13,000 in total, usually paid module by module. Imperial's online MSc Machine Learning and Data Science is about £39,900 for 2026 entry. Fees change each year, so confirm the current figure on the university's own page before applying. ### Is a UK data science bootcamp worth it? It can be, particularly for career switchers who need structure. Government-funded Skills Bootcamps are free for eligible adults in England, while private bootcamps such as Le Wagon, Makers and Northcoders charge fees or offer income-share arrangements. Compare the syllabus, total cost and outcome data that counts everyone who enrolled. ### Does an online UK degree lead to a Graduate Route visa? Generally not if you study from outside the UK. The Graduate Route is for eligible students who completed a UK degree while studying in the UK, and immigration rules change regularly. Check current requirements on GOV.UK before relying on any course for visa purposes. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/best-data-science-ai-courses-uk-2026/). Cite with the title, "1stepGrow" and a link._ --- # Best Data Science Courses in Europe for 2026: Degrees to Bootcamps > The best data science courses in Europe for 2026, from low-fee master's degrees to ETH Zurich certificates and bootcamps, and why nationality changes the price. - **Author:** Ravi Pradhan — AI Expert - 11+ Yrs Experience (https://www.1stepgrow.com/authors/ravi-pradhan/) - **Published:** Sep 1, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 10 min - **Canonical URL:** https://www.1stepgrow.com/articles/best-data-science-ai-courses-europe-2026/ ## Key takeaways - For EU/EEA citizens, public master's degrees such as EIT Digital, TU Munich and the University of Amsterdam offer serious curricula at very low tuition. - Non-EU students should check fees carefully: EIT Digital and TU Munich both charge non-EU tuition, and Dutch institutional fees are well above the statutory rate. - Working professionals get more from part-time routes such as ETH Zurich's CAS programmes, part-time bootcamps or online certificates than from quitting to study full time. - Germany's Bildungsgutschein and France's CPF can fund approved courses for eligible residents, so check a provider's approval status before you pay. The best data science courses in Europe for most people in 2026 are public master's degrees: the EIT Digital Master School's Data Science double degree, TU Munich's M.Sc. Data Engineering and Analytics and the University of Amsterdam's one-year track. The catch is nationality. EIT Digital is free for EU/EEA citizens, for example, but costs about €16,000 a year for everyone else. This guide is for graduates, working professionals and non-EU applicants choosing between a degree, part-time continuing education and a bootcamp. It ranks eight options, from ETH Zurich certificates to Le Wagon and TU Delft's free edX course. It then covers what changes the real cost of each: ECTS credit, the teaching language, funding such as Germany's Bildungsgutschein and France's CPF, and visa rules. ## How we compared these courses We compared each programme's published curriculum, fees, format and credential using the providers' official pages, as they stood in September 2026. We did not survey students, interview alumni or audit job outcomes, and we have left out placement claims because providers do not publish them in a comparable form. The ratings are our editorial scores reflecting the criteria below, not user reviews. - **Curriculum depth.** Does the programme go beyond introductory Python into statistics, machine learning, deep learning, data engineering and deployment? Does it address generative AI and responsible AI in a way that reflects European regulation? - **Format.** Full-time, part-time, online or blended, and how realistic the weekly workload is. - **Credential recognition.** Accredited degree, ECTS-bearing certificate, national certification such as a French RNCP title, or private certificate. - **Total cost.** Tuition for EU/EEA and non-EU students, plus the living and opportunity costs a programme implies. - **Support.** Teaching contact, mentoring, projects and career services where they are published. - **Fit for working professionals.** Whether you can realistically combine the course with a job. As an AI practitioner who has hired and mentored data scientists across EdTech, financial services and technology, I weighted curriculum depth and credential recognition most heavily. They are what hiring managers can verify. ### 1. Data Science (double-degree master's) Two universities, two countries, two degrees, plus an innovation and entrepreneurship minor. - **Provider:** EIT Digital Master School - **Price:** Tuition-free for EU/EEA students; approx. €16,000 per year for non-EU students (scholarships available) - **Duration:** 2 years full-time (120 ECTS) - **Level:** Intermediate - **1stepGrow score:** 4.6/5 - **Best for:** Graduates who want a rigorous, internationally mobile master's with a business edge - **Website:** https://masterschool.eitdigital.eu/data-science **Pros** - Study at an entry university in one country and an exit university in another, with a degree from each - Technical data science core combined with an innovation and entrepreneurship minor - Universities across the programme include institutions such as Aalto University, Politecnico di Milano and ELTE **Cons** - Non-EU tuition is significant unless you win a merit scholarship - Moving countries after year one adds cost and admin **Verdict:** The strongest overall pick for a recent graduate who wants depth, mobility and a recognised European degree. EU/EEA applicants get exceptional value. ### 2. M.Sc. Data Engineering and Analytics An engineering-heavy, English-taught master's at one of Germany's leading technical universities. - **Provider:** Technical University of Munich (TUM) - **Price:** Semester fee only for EU/EEA students; non-EU students typically pay approx. €4,000–€6,000 per semester - **Duration:** 4 semesters full-time (120 ECTS) - **Level:** Advanced - **1stepGrow score:** 4.5/5 - **Best for:** Computer science graduates aiming for data engineering or ML engineering roles - **Website:** https://www.cit.tum.de/en/cit/studies/degree-programs/master-data-engineering-and-analytics/ **Pros** - Taught in English with a strong focus on large-scale data processing and analysis - Munich offers a large job market in automotive, insurance and technology - Very low cost for EU/EEA students, who pay only the semester fee **Cons** - Requires a bachelor's in informatics or an equivalent technical degree - Tuition for non-EU students since winter semester 2024/25 has changed the value calculation **Verdict:** Excellent for technically strong applicants who want an engineering-first data degree. Non-EU students should confirm the current fee on the programme page. ### 3. MSc Information Studies: Data Science track A one-year, English-taught master's in the middle of one of Europe's busiest tech cities. - **Provider:** University of Amsterdam - **Price:** Approx. €2,700 statutory fee (2026–27) for eligible EU/EEA students; higher institutional fee for non-EU students - **Duration:** 12 months full-time (60 ECTS), part-time option available - **Level:** Intermediate - **1stepGrow score:** 4.4/5 - **Best for:** Graduates who want a fast, recognised master's and access to Amsterdam's international job market - **Website:** https://www.uva.nl/en/programmes/masters/information-studies-data-science/data-science.html **Pros** - Only one year, which reduces living costs and time out of work - English-taught, and English is widely used in Amsterdam workplaces - Part-time mode available **Cons** - One year leaves less room for depth than a two-year programme - Amsterdam housing is expensive and hard to find **Verdict:** The quickest route to a full European master's on this list. It suits people who already have some technical background and want to be job-hunting within a year. ### 4. Erasmus Mundus Joint Master's in data science and AI Multi-country master's degrees, with EU-funded scholarships for top-ranked applicants. - **Provider:** Erasmus+ / European Education and Culture Executive Agency - **Price:** Varies by programme; full scholarships covering tuition, travel and living allowance for top applicants - **Duration:** 1–2 years full-time (varies by programme) - **Level:** Intermediate - **1stepGrow score:** 4.3/5 - **Best for:** Strong applicants from anywhere in the world who want to study in several countries on a scholarship - **Website:** https://www.eacea.ec.europa.eu/scholarships/erasmus-mundus-catalogue_en **Pros** - Each programme is run by at least three institutions in at least three countries - Scholarships can remove most of the cost for successful applicants - The catalogue includes several data science and AI-related programmes **Cons** - Scholarships are highly competitive, and admission alone does not guarantee funding - Each programme has its own curriculum, deadlines and requirements, so research takes time **Verdict:** Worth an application from any strong candidate. Use the official catalogue to find current data science and AI programmes, and apply early, usually between October and January. ### 5. CAS ETH in AI, Data and Machine Learning A short, credit-bearing certificate from one of Europe's top technical universities, built for managers. - **Provider:** ETH Zurich (School for Continuing Education) - **Price:** Approx. CHF 8,500 (Swiss francs) - **Duration:** 14 weeks, around 300 hours of study (12 ECTS) - **Level:** Beginner - **1stepGrow score:** 4.2/5 - **Best for:** Experienced managers who need to lead AI and data initiatives without a computer science degree - **Website:** https://sce.ethz.ch/en/programmes-and-courses/search-current-courses/cas/cas-eth-dml.html **Pros** - Carries 12 ECTS and the ETH Zurich name - Block or blended format designed around a working schedule - Taught in English **Cons** - Requires a master's-level qualification and several years of management experience - Designed for AI literacy and leadership, not hands-on data scientist training **Verdict:** A strong choice for senior professionals in Switzerland and nearby who need credible AI fluency. Analysts who want hands-on statistics should look at ETH's CAS in Applied Statistical Data Science instead. ### 6. Data Science & AI Bootcamp An intensive, project-based bootcamp running in cities across Europe and online. - **Provider:** Le Wagon - **Price:** Approx. €7,000–€9,000 depending on city and format - **Duration:** About 9 weeks full-time; longer part-time option - **Level:** Beginner - **1stepGrow score:** 4.1/5 - **Best for:** Career changers who want structure, a cohort and a portfolio in a few months - **Website:** https://www.lewagon.com/data-science-course **Pros** - Curriculum runs from Python and SQL through machine learning, deep learning and ML engineering with Docker and cloud deployment - Campuses in cities such as Paris, Berlin, Amsterdam and Lisbon, plus online cohorts - Ends with a team project you can show employers **Cons** - Nine weeks is a very fast pace for complete beginners - A bootcamp certificate is not an accredited degree **Verdict:** One of the better-structured bootcamps in Europe for motivated career changers. Complete the prep work seriously and treat the final project as your portfolio centrepiece. ### 7. Data Scientist Machine Learning path An online, mentor-supported path leading to a French state-recognised RNCP level 7 certification. - **Provider:** OpenClassrooms - **Price:** Approx. €5,000–€9,000 depending on format; CPF or apprenticeship funding possible in France - **Duration:** Varies by format and pace - **Level:** Intermediate - **1stepGrow score:** 4/5 - **Best for:** Learners in France who want a recognised master's-level certification while studying online - **Website:** https://openclassrooms.com/en/paths/1047-data-scientist-machine-learning **Pros** - Leads to an RNCP level 7 certification recognised by the French state - Project-based, with around a dozen professional projects and one-to-one mentoring - Work-study (apprenticeship) route available, where an employer covers the cost **Cons** - Non-native speakers are expected to have at least B2 French - Self-paced online study demands strong discipline **Verdict:** A practical option for French-speaking learners who need a formal credential without attending campus. Outside France, the RNCP title is less widely recognised. ### 8. AI Skills: Basic and Advanced Techniques in Machine Learning A free-to-access professional certificate from a leading Dutch technical university. - **Provider:** TU Delft (via edX) - **Price:** Free to access; optional certificate fees vary - **Duration:** About 12 weeks self-paced, 5–7 hours per week - **Level:** Beginner - **1stepGrow score:** 3.9/5 - **Best for:** Engineers and self-directed learners who want a low-risk introduction to machine learning - **Website:** https://learningforlife.tudelft.nl/ai-skills-basic-and-advanced-techniques-in-machine-learning/ **Pros** - Two courses covering supervised, unsupervised, deep and reinforcement learning - Hands-on Python and scikit-learn work, ending in a machine learning pipeline project - No cost to start **Cons** - No live teaching or career support - An introductory certificate, not a degree or ECTS-bearing qualification **Verdict:** The best free starting point on this list. Use it to test your interest before committing to a degree or bootcamp. ## How do the eight courses compare on cost and format? The three master's degrees are cheapest for EU/EEA citizens but take one to two years; the certificates and bootcamps take weeks rather than years and carry little or no ECTS credit. | Course | Format | Duration | Indicative fee | Best for | |---|---|---|---|---| | EIT Digital Master School, Data Science | Full-time, two countries | 2 years (120 ECTS) | Free for EU/EEA; approx. €16,000/yr non-EU | Mobile graduates wanting a double degree | | TU Munich, M.Sc. Data Engineering and Analytics | Full-time, on campus | 4 semesters (120 ECTS) | Semester fee (EU/EEA); approx. €4,000–€6,000/semester non-EU | CS graduates targeting data engineering | | University of Amsterdam, MSc Data Science | Full-time or part-time | 12 months (60 ECTS) | Approx. €2,700 statutory fee; higher for non-EU | A fast, recognised master's | | Erasmus Mundus joint master's | Full-time, multi-country | 1–2 years | Varies; scholarships available | Strong applicants seeking funding | | ETH Zurich, CAS in AI, Data and ML | Part-time, block or blended | 14 weeks (12 ECTS) | Approx. CHF 8,500 | Managers leading AI work | | Le Wagon, Data Science & AI Bootcamp | Full-time or part-time, campus or online | About 9 weeks full-time | Approx. €7,000–€9,000 | Career changers | | OpenClassrooms, Data Scientist ML path | Online, mentored | Varies | Approx. €5,000–€9,000 | French-speaking online learners | | TU Delft, AI Skills (edX) | Online, self-paced | About 12 weeks | Free to access | Low-risk first step | As of September 2026, all prices are indicative and change by intake, nationality and currency, so confirm the current fee on the provider's official page before applying. ## How do you choose among the best data science courses in Europe? ### Degree, continuing education or bootcamp? The three main routes solve different problems. A **master's degree** is the right tool if you are early in your career, want to work for large employers or research-led teams, or need a recognised qualification for immigration purposes. It is also the slowest route: one to two years of full-time study, plus an application cycle that for many European programmes opens in the autumn or winter before a September start. **Continuing education**, such as ETH Zurich's CAS programmes, suits people who already have a degree and a career and need to add data or AI capability without stopping work. You get university-level teaching and ECTS credits, but not a full qualification in data science. A **bootcamp** is built for speed. In roughly nine to twenty-four weeks you leave with practical skills and a portfolio project. That can be enough for analyst or junior roles if you bring relevant domain experience, but it does not replace a degree where employers or visa rules require one. ### What else to weigh **Start with your residency status.** It shapes cost more than anything else. EU/EEA citizens can often study a full master's for the price of a semester contribution or a statutory fee. Non-EU applicants face tuition at EIT Digital, TU Munich and Dutch universities, so for them the gap between a degree and a part-time certificate is much smaller than it looks. **Understand the credit.** ECTS makes European programmes comparable. [Sixty ECTS is the equivalent of a full year of study](https://education.ec.europa.eu/education-levels/higher-education/inclusive-and-connected-higher-education/european-credit-transfer-and-accumulation-system), and one credit usually represents about 25 to 30 hours of work. That tells you ETH Zurich's 12-ECTS CAS is roughly a fifth of a full-time year, while TU Munich's 120-ECTS master's is two full years. Bootcamps and most online certificates carry no ECTS at all. That is fine for building skills, but it matters if you later want credit towards a degree or need a formal qualification. **Check the language, all the way through.** Many programmes are fully English-taught, including TUM's Data Engineering and Analytics, the University of Amsterdam's Data Science track and EIT Digital. Others mix languages, and some French online diplomas expect at least B2 French. Even on an English-taught degree, learning the local language widens your job options in Munich, Paris or Madrid. **Use funding you are entitled to.** In Germany, the [Bildungsgutschein](https://www.arbeitsagentur.de/karriere-und-weiterbildung/bildungsgutschein) is a training voucher that the employment agency or Jobcenter can issue to eligible job seekers and people at risk of unemployment. It covers AZAV-certified courses, and Ironhack's online Data Science & Machine Learning bootcamp in Germany (listed at about €8,500) is one that advertises eligibility. In France, the CPF personal training account can fund certified courses such as RNCP titles from OpenClassrooms and Liora (formerly DataScientest), though learners now contribute part of the cost. Erasmus Mundus scholarships are open to applicants worldwide. **Think about regulation as a skill.** The EU AI Act and GDPR mean European employers value data people who understand documentation, risk assessment and privacy by design. A programme that treats responsible AI as a real module rather than a single lecture is better preparation for work in Europe. **For international students, keep visas in view.** A degree programme with physical attendance is usually needed for a student residence permit, and online-only study generally does not qualify. After graduating, many EU countries allow a job-search period, and the [EU Blue Card](https://home-affairs.ec.europa.eu/policies/migration-and-asylum/eu-immigration-portal/eu-blue-card_en) is a common route for qualifying skilled roles. By contrast, Ireland, Denmark and Switzerland run their own national systems. Rules change, however, so confirm them on official government sites. **Look beyond the eight.** Other credible options include KU Leuven's Master of Artificial Intelligence in Belgium, data science and analytics MSc programmes at Trinity College Dublin and University College Dublin, Ironhack campuses in Madrid, Barcelona, Berlin, Paris, Lisbon and Amsterdam, and European business-school courses such as HEC Paris's on Coursera. For a full regional overview, see our [Europe hub](https://www.1stepgrow.com/regions/europe/). ### Questions to ask before you enrol Whichever route you pick, get clear answers to these before you pay a deposit: - What is the total cost for someone of my nationality, including semester fees, materials and any exam or certification fees? - Is every module taught in English, and are any assessments in the local language? - How many ECTS credits does the programme carry, and will another university recognise them? - What are the weekly contact hours, and how many hours of self-study are expected on top? - For bootcamps, is the course AZAV-certified in Germany or CPF-eligible in France, and what exactly does career support include? - When was the curriculum last updated, and does it cover deployment and responsible AI as well as modelling? A provider that answers these clearly and in writing is usually one that runs its programme well. ## Which course should you choose? **You are a recent EU/EEA graduate with a quantitative degree.** Apply to EIT Digital's Data Science programme and TU Munich, and add the University of Amsterdam if you want to finish in a year. At these fee levels, a full European master's is hard to beat. **You are a non-EU graduate on a tight budget.** Put Erasmus Mundus scholarships at the top of your list, and apply for EIT Digital's merit scholarships. Compare the total cost of TUM's non-EU tuition with Dutch institutional fees before deciding, including rent in Munich and Amsterdam. **You are a manager in Switzerland or southern Germany.** ETH Zurich's CAS in AI, Data and Machine Learning gives you a credible, ECTS-bearing credential and AI fluency without leaving your job. If you want hands-on skills, look at ETH's CAS in Applied Statistical Data Science. **You are changing careers in Berlin, Paris, Amsterdam or Lisbon.** A Le Wagon or Ironhack bootcamp gives you structure and a portfolio in weeks. If you are unemployed in Germany, ask your advisor about a Bildungsgutschein before paying anything yourself. **You live in France and want a formal credential online.** OpenClassrooms' Data Scientist path, or an equivalent RNCP certification from Liora, can be funded through CPF or an apprenticeship. **You are a software engineer moving into machine learning.** If you can take two years out, TU Munich's Data Engineering and Analytics programme builds directly on a computer science degree. If you cannot, TU Delft's free certificate plus your own deployment projects will close the gap faster than a general data science course. So will the ML engineering modules of a part-time bootcamp. **You are not sure data science is for you.** Start with TU Delft's free AI Skills certificate on edX. Spend twelve weeks on it before spending thousands of euros. ## Related reading - [Data science and AI in Europe: courses, careers and funding](https://www.1stepgrow.com/regions/europe/) - [Data scientist roadmap 2026](https://www.1stepgrow.com/articles/data-scientist-roadmap-2026) - [Is a data science bootcamp worth it?](https://www.1stepgrow.com/articles/is-a-data-science-bootcamp-worth-it) - [Best AI courses in 2026](https://www.1stepgrow.com/articles/best-ai-courses-2026) ## Frequently asked questions ### What is the best data science course in Europe? For most recent graduates, the EIT Digital Master School's Data Science double degree is the strongest all-round choice, followed closely by TU Munich's M.Sc. Data Engineering and Analytics and the University of Amsterdam's one-year MSc Data Science track. Working professionals are usually better served by ETH Zurich's CAS programmes, a part-time bootcamp such as Le Wagon, or an online certificate. ### Are data science master's degrees free in Europe? Some are close to free for EU/EEA citizens. EIT Digital charges them no tuition, and German public universities such as TUM charge only a semester fee. Non-EU students increasingly pay tuition, for example at EIT Digital and TUM, so always check the fee for your nationality. ### Can I study data science in Europe in English? Yes. Many master's programmes, including those at TUM, the University of Amsterdam and EIT Digital, are taught entirely in English, as are ETH Zurich's CAS in AI, Data and Machine Learning and most bootcamps. Some French online diplomas expect French. ### Can I get a bootcamp funded in Germany? Eligible job seekers can apply for a Bildungsgutschein through the employment agency or Jobcenter, which can cover AZAV-certified courses. Some data bootcamps, including Ironhack's online programme in Germany, advertise Bildungsgutschein eligibility. Approval depends on your individual situation. ### Is a bootcamp or a master's better for getting a data job in Europe? A master's is the safer choice if you are early in your career, want roles at larger employers or need a qualification for a visa. A bootcamp can work well for career changers with relevant experience who need a portfolio quickly. Many people start with a free course such as TU Delft's to find out which route suits them. ### How much does a data science course in Europe cost? Costs range from free online courses, to about €2,700 a year for an eligible EU student at a Dutch university, to roughly €7,000 to €9,000 for a bootcamp and CHF 8,500 for an ETH Zurich CAS. Non-EU master's fees can be much higher. All figures are indicative, so confirm them on the provider's page. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/best-data-science-ai-courses-europe-2026/). Cite with the title, "1stepGrow" and a link._ --- # Best Data Science Courses in Dubai and the UAE for 2026 > The best data science courses in Dubai and the UAE for 2026: MBZUAI, UK branch campuses, bootcamps and online options, with AED fees and who each suits. - **Author:** Ravi Pradhan — AI Expert - 11+ Yrs Experience (https://www.1stepgrow.com/authors/ravi-pradhan/) - **Published:** Aug 31, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 13 min - **Canonical URL:** https://www.1stepgrow.com/articles/best-data-science-ai-courses-uae-2026/ ## Key takeaways - MBZUAI's MSc in Machine Learning is the strongest option for full-time learners aiming at AI research, and admitted students have historically been fully funded (confirm current terms). - For a recognised one-year taught degree in Dubai, the University of Birmingham Dubai and Heriot-Watt University Dubai lead, at roughly AED 95,000–155,000 in tuition. - Most university MSc programmes here are full-time, so working professionals should look at MBZUAI's part-time Master in Applied AI, part-time bootcamps or online certificates. - Check how a degree is recognised, because CAA accreditation and KHDA permits serve different purposes, before you pay a deposit. The best data science courses in Dubai and the wider UAE depend on your goal. For AI research, MBZUAI's MSc in Machine Learning in Abu Dhabi stands out, and its admitted students have historically been fully funded. For a one-year taught degree in Dubai itself, the University of Birmingham Dubai and Heriot-Watt University Dubai lead, at roughly AED 95,000–155,000. The harder problem for most readers is keeping a job, because nearly every one-year MSc here is full-time. If you need job-ready skills quickly and cheaply, Le Wagon Dubai or the IBM Data Science Professional Certificate will get you further per dirham. This guide is for graduates, working professionals and career changers in the Emirates. It ranks eight programmes with indicative AED fees, formats and the trade-offs their marketing pages skip, explains how CAA accreditation differs from a KHDA permit, and ends with a pick for each situation. Fees change every intake, so treat the figures as a starting point and confirm them before you apply. ## How we compared these courses I have spent more than 11 years working in AI and data across EdTech, financial services and technology, and the question I hear most often from people in the Gulf is not which course is best, but which one is worth their time given their job, visa and budget. So we compared these programmes on six criteria that map to that decision: - **Curriculum depth.** Does the programme cover statistics, machine learning, deep learning and data engineering properly, and does it touch generative AI and deployment? Does it end in a substantial project or thesis? - **Format.** Full-time, part-time, evening and weekend, or online, and what that means for someone keeping a job in Dubai or Abu Dhabi. - **Credential recognition.** Who awards the qualification, how it is recognised in the UAE (CAA accreditation, KHDA permit or neither) and how familiar local employers are with it. - **Total cost.** Tuition in AED, plus the costs people forget: application fees, deposits, visa and living expenses, and time out of work. We also looked at scholarships and instalment options. - **Support.** Supervision, instructor access, cohort size, careers support and how much of the learning is autograded versus reviewed by a person. - **Fit for working professionals.** Whether someone in a full-time UAE job can realistically complete the programme. This is a comparison of published curricula, fees, formats and credentials on each provider's official pages as of September 2026. We did not survey students or rely on placement claims. The ratings are our editorial scores reflecting how well each programme serves the learner it is designed for, not a ranking of institutions overall. ### 1. Master of Science in Machine Learning A research-led AI master's from a graduate university dedicated entirely to artificial intelligence. - **Provider:** Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), Abu Dhabi - **Price:** Fully funded for admitted full-time students historically (check current terms) - **Duration:** 2 years full-time - **Level:** Advanced - **1stepGrow score:** 4.8/5 - **Best for:** Strong STEM graduates aiming for AI research, a PhD or ML engineering at research-heavy employers - **Website:** https://mbzuai.ac.ae/study/msc-programs/master-of-science-in-machine-learning/ **Pros** - Coursework in year one and a supervised research thesis in year two, with a course plan tailored to each student - Admitted full-time students have historically received full tuition coverage plus a monthly stipend and accommodation support - Abu Dhabi location puts you close to the UAE's AI research and infrastructure employers **Cons** - Highly selective: expects a STEM bachelor's, strong grades, solid maths and at least intermediate programming - Full-time and research-focused, so it cannot be combined with a full-time job **Verdict:** The best AI master's in the UAE if you can study full-time and want depth rather than a quick credential. Working professionals should look at MBZUAI's separate, fee-based Master in Applied AI instead. ### 2. MSc Data Science A UK Russell Group master's taught in Dubai, with teaching shared between Dubai and Birmingham staff. - **Provider:** University of Birmingham Dubai - **Price:** approx. AED 145,000–155,000 (tuition, one year) - **Duration:** 12 months full-time - **Level:** Intermediate - **1stepGrow score:** 4.6/5 - **Best for:** Graduates and early-career professionals who want a well-known UK degree without leaving Dubai - **Website:** https://www.birmingham.ac.uk/dubai/study/postgraduate/subjects/computer-science-courses/data-science-msc **Pros** - Covers statistical methods, machine learning, programming, data management and visualisation - Includes a substantial 60-credit research project - The university states the programme is accredited by the UAE's federal higher-education ministry, and related MScs in financial and health data science also run in Dubai **Cons** - The most expensive taught MSc on this list - The full-time, one-year format is hard to combine with a demanding job **Verdict:** Our top taught-degree pick for Dubai: rigorous, well recognised and genuinely focused on data science rather than a rebadged business course. Budget carefully, because the fee is at the top of the market. ### 3. MSc Data Science A long-established UK university in Dubai International Academic City offering a one-year data science MSc. - **Provider:** Heriot-Watt University Dubai - **Price:** approx. AED 95,000–105,000 (tuition, one year) - **Duration:** 12 months full-time - **Level:** Intermediate - **1stepGrow score:** 4.5/5 - **Best for:** Students who want a UK degree in Dubai at a lower fee than the most expensive branch campuses - **Website:** https://www.hw.ac.uk/dubai/study/postgraduate/data-science **Pros** - Heriot-Watt has operated in Dubai for many years and is familiar to UAE employers - Awards a Heriot-Watt University degree - Lower tuition than several comparable branch-campus MSc programmes **Cons** - Check the current module list carefully, as coverage of generative AI and MLOps varies by year - The full-time format limits options for people in full-time work **Verdict:** A solid, better-value alternative to Birmingham Dubai with a similar credential profile. Compare the two module lists side by side before choosing. ### 4. MSc in Computational Data Science A thesis-based data science master's from one of the UAE's leading research universities. - **Provider:** Khalifa University, Abu Dhabi - **Price:** approx. AED 150,000 list tuition; merit scholarships can waive fees (check current terms) - **Duration:** Typically 1–2 years full-time (thesis-based) - **Level:** Advanced - **1stepGrow score:** 4.5/5 - **Best for:** Research-minded graduates in engineering, maths or computer science who want a degree from a UAE federal-level research university - **Website:** https://www.ku.ac.ae/program/msc-in-computational-data-science **Pros** - Combines computer science algorithms with statistical methods for large-scale data - Includes an independent research thesis supervised by faculty - Competitive merit-based scholarships can cover tuition for strong applicants **Cons** - Admission has required a GRE quantitative score, a hurdle most other programmes here skip - The research focus means less emphasis on business-facing analytics and deployment **Verdict:** A strong research alternative to MBZUAI, particularly if your interests sit between applied maths and data science. Apply early and ask explicitly how scholarship tiers work. ### 5. Master of Science in Machine Learning A US-style machine learning master's from AUS's Department of Computer Science and Engineering, within commuting distance of Dubai. - **Provider:** American University of Sharjah - **Price:** approx. AED 150,000–170,000 total tuition (billed per credit hour) - **Duration:** Minimum 30 credit hours; typically 1.5–2 years - **Level:** Advanced - **1stepGrow score:** 4.4/5 - **Best for:** Computer science and engineering graduates who want a technical ML degree on a US-style credit system - **Website:** https://www.aus.edu/cen/department-of-computer-science-and-engineering/master-of-science-in-machine-learning **Pros** - Focused on machine learning rather than general analytics - Designed to prepare students for advanced industry roles or doctoral study - AUS is also launching an MS in Data Analytics Engineering, designed with working professionals in mind, from Fall 2026 **Cons** - Expects a computer science or computer engineering background; other applicants are considered case by case - The Sharjah campus means a commute for Dubai-based professionals **Verdict:** The most engineering-heavy option outside Abu Dhabi. If you are in full-time work, compare it with AUS's newer data analytics engineering master's, which is scheduled for people in employment. ### 6. MSc Data Science and Artificial Intelligence A hands-on, lab-heavy MSc on the Dubai Knowledge Park campus, covering data engineering through to generative models. - **Provider:** Middlesex University Dubai - **Price:** approx. AED 75,000–85,000 (tuition, one year) - **Duration:** 1 year full-time - **Level:** Intermediate - **1stepGrow score:** 4.3/5 - **Best for:** Graduates from numerate disciplines who want a UK degree at the lowest branch-campus fee on this list - **Website:** https://www.mdx.ac.ae/courses/msc-data-science-and-artificial-intelligence **Pros** - Lowest tuition among the UK branch-campus MSc programmes we compared - Practical emphasis, with significant supervised lab time - Curriculum spans databases, statistics, machine learning, NLP and generative models **Cons** - Less research depth than MBZUAI, Khalifa University or Birmingham Dubai - Full-time, one-year format **Verdict:** The value pick among Dubai's UK campuses. A good fit if you want a recognised master's and practical skills without paying top-of-market fees. ### 7. Data Science & AI Bootcamp An intensive, project-based bootcamp run from a Dubai campus, full-time, part-time or online. - **Provider:** Le Wagon Dubai - **Price:** approx. AED 25,000–35,000 (instalments available) - **Duration:** About 2 months full-time or 6 months part-time - **Level:** Beginner - **1stepGrow score:** 4.2/5 - **Best for:** Career changers and analysts who want practical Python and machine learning skills fast - **Website:** https://www.lewagon.com/dubai **Pros** - The part-time track runs on nights and weekends, so you can keep your job - Project-based learning with live instructors and a cohort - Instalment plans and scholarships, with further funding routes listed on its Dubai page **Cons** - Not a degree, so employers will judge you on your portfolio - The full-time track demands around 40 hours a week **Verdict:** The best bootcamp option physically in Dubai. Worth it if you will put in the hours and build real projects; not a substitute for a master's if you want research roles. ### 8. IBM Data Science Professional Certificate The most widely recognised low-cost entry credential in data science, fully online. - **Provider:** IBM via Coursera - **Price:** approx. AED 180–220 per month (Coursera subscription) - **Duration:** Around 4–6 months at about 10 hours a week - **Level:** Beginner - **1stepGrow score:** 4.1/5 - **Best for:** Complete beginners testing whether data science suits them before paying for a bootcamp or degree - **Website:** https://www.coursera.org/professional-certificates/ibm-data-science **Pros** - Very low total cost, because you pay only while subscribed - Self-paced, so it fits around any work schedule - Covers Python, SQL, data analysis, visualisation and introductory machine learning **Cons** - Autograded work and forums rather than personal mentoring - Too introductory on its own to win data scientist roles **Verdict:** The cheapest sensible first step. Finish it, build one project on local data, then decide whether a bootcamp or degree is worth the money. ## How do the eight UAE programmes compare? Taught MSc fees in Dubai range from about AED 75,000 at Middlesex to AED 155,000 at Birmingham, the research degrees in Abu Dhabi can be funded, and the bootcamp and online certificate cost a fraction of either. | Course | Format | Duration | Indicative fee | Best for | |---|---|---|---|---| | MBZUAI: MSc in Machine Learning | Full-time, on campus (Abu Dhabi) | 2 years | Fully funded for admitted students historically (check terms) | Future AI researchers and ML engineers | | University of Birmingham Dubai: MSc Data Science | Full-time, on campus (Dubai) | 12 months | approx. AED 145,000–155,000 | A well-known UK degree in Dubai | | Heriot-Watt University Dubai: MSc Data Science | Full-time, on campus (Dubai) | 12 months | approx. AED 95,000–105,000 | A UK degree at a mid-market fee | | Khalifa University: MSc in Computational Data Science | Full-time, thesis-based (Abu Dhabi) | Typically 1–2 years | approx. AED 150,000 list; scholarships available | Research-minded quantitative graduates | | American University of Sharjah: MS in Machine Learning | On campus (Sharjah), credit-based | Typically 1.5–2 years | approx. AED 150,000–170,000 total | CS and engineering graduates | | Middlesex University Dubai: MSc Data Science and AI | Full-time, on campus (Dubai) | 1 year | approx. AED 75,000–85,000 | A UK degree on a tighter budget | | Le Wagon Dubai: Data Science & AI Bootcamp | Full-time, part-time or online | 2 or 6 months | approx. AED 25,000–35,000 | Career changers who want practical skills fast | | IBM Data Science Professional Certificate (Coursera) | Online, self-paced | 4–6 months | approx. AED 180–220 per month | Beginners testing the field | As of September 2026, all prices are indicative and based on providers' published information. Confirm them on the provider's official page before you apply. ## How should you weigh up the best data science courses in Dubai and Abu Dhabi? **Start with how the degree is recognised.** UAE higher education runs on two tracks. Universities licensed at federal level have their programmes accredited through the Commission for Academic Accreditation (CAA), which publishes a [list of accredited programmes](https://www.caa.ae/Pages/Programs/All.aspx). Branch campuses in Dubai's free zones, such as Dubai International Academic City and Dubai Knowledge Park, operate under permits from the [Knowledge and Human Development Authority (KHDA)](https://web.khda.gov.ae/en/Guides/Students/FAQs-on-universities-in-Dubai) and award their home-country degree. Some branch-campus programmes also carry CAA accreditation. For most private-sector jobs, however, either route is fine. If you might want a federal government role, a licensed profession or a PhD elsewhere in the region, ask the university in writing how its degree is recognised and whether equivalency is needed. **Count the total cost, not the headline fee.** A one-year MSc at a Dubai branch campus is a tuition bill of roughly AED 75,000–155,000 before rent, a student visa if you need one, and a year of lost or reduced salary. That is why the two most interesting options on price are at opposite ends of the list: MBZUAI, where admitted full-time students have historically been fully funded, and online certificates that cost a few hundred dirhams a month. Khalifa University's merit scholarships can also change the maths considerably for strong applicants. **Match the timetable to your working week.** Since 2022, most UAE employers have worked Monday to Friday with a Saturday–Sunday weekend. Yet nearly all the one-year MSc programmes above are full-time. If you are keeping your job, the realistic choices are MBZUAI's Master in Applied Artificial Intelligence, a fee-based programme of about two years with evening and weekend classes (MBZUAI lists the total tuition at approx. AED 170,000), Le Wagon's part-time bootcamp, or online study. **Check the entry requirements early.** The research programmes are selective. MBZUAI expects a STEM bachelor's with strong grades and solid maths, and Khalifa University's data science MSc has required a GRE quantitative score. Branch campuses are generally more flexible and accept graduates from any numerate discipline, sometimes with relevant work experience in lieu of a technical degree. **Think about language and local data.** Almost every programme here is taught in English, which matches the working language of UAE tech teams. If you want to work in government or on Arabic-language AI, choose projects that use Arabic text or local datasets. It is the easiest way to stand out from applicants with the same certificate. **Ask five questions before you pay a deposit.** Admissions teams in Dubai are used to these, and a clear written answer to each will save you from most bad decisions: 1. How is this specific programme recognised in the UAE: CAA accreditation, a KHDA permit or both, and is any equivalency step needed for my purpose? 2. What is the full cost, including application fees, deposits, VAT, visa and any lab or graduation charges? 3. Which modules are compulsory this intake, and when was the curriculum last updated to cover generative AI and model deployment? 4. What does the final project or thesis involve, and can it be done with a UAE employer's data? 5. If I fall behind or change jobs, can I defer, switch to part-time or pause without losing fees? **Build proof alongside the credential.** Whatever you choose, finish with work you can show. A project on public Dubai or Abu Dhabi data, such as transport, property or tourism datasets, tells a local hiring manager far more than a transcript does, and it gives you something concrete to discuss in interviews. **Other options worth knowing.** The University of Dubai offers an MSc in Data Science, and Canadian University Dubai has an MSc in Artificial Intelligence. General Assembly runs online data courses accessible from the UAE. Google's data analytics certificates and Microsoft's Azure data and AI certifications on Coursera and Microsoft Learn are sensible add-ons if your target employer uses those platforms. The University of Birmingham Dubai also runs an MSc in Financial Data Science, which is worth a look if you are aiming at DIFC. ## Which course should you choose? **You are a recent STEM graduate who wants to do AI research or a PhD.** Apply to MBZUAI's MSc in Machine Learning first, and to Khalifa University's MSc in Computational Data Science as a second research option. Both are selective; both can dramatically reduce your costs if you are admitted with funding. **You live in Dubai, work full-time and want a master's degree.** Most branch-campus MSc programmes will force you to choose between the degree and your job. MBZUAI's Master in Applied AI is designed for exactly your situation, with evening and weekend classes. If your employer will release you for a year, the University of Birmingham Dubai or Heriot-Watt University Dubai are the strongest taught degrees. **You work in banking or fintech in DIFC.** The University of Birmingham Dubai's MSc Data Science, or its MSc in Financial Data Science, maps most directly to credit risk, fraud and trading analytics. If a year out is not possible, build ML skills part-time through Le Wagon and apply them to problems at your current employer. **You are changing careers on a budget.** Start with the IBM Data Science Professional Certificate. If you finish it and still enjoy the work, Le Wagon Dubai's part-time bootcamp is the logical next step. Our analysis of [whether a data science bootcamp is worth it](https://www.1stepgrow.com/articles/is-a-data-science-bootcamp-worth-it) covers how to judge the return. **You want a UK degree without top-of-market fees.** Middlesex University Dubai's MSc Data Science and Artificial Intelligence is the value pick, and Heriot-Watt University Dubai sits in the middle of the market. **You manage a team and need to understand AI rather than build it.** A research MSc is overkill. Look at MBZUAI's applied master's if you want a full degree, or combine a short online AI course with a structured reading plan from our [best AI courses guide](https://www.1stepgrow.com/articles/best-ai-courses-2026). ## Related reading For the local job market, accreditation rules and in-demand skills, see our [UAE tech courses and careers hub](https://www.1stepgrow.com/regions/uae/). If you are mapping out the full skill path, start with the [data scientist roadmap for 2026](https://www.1stepgrow.com/articles/data-scientist-roadmap-2026). For a global view of AI programmes, read our guide to the [best AI courses in 2026](https://www.1stepgrow.com/articles/best-ai-courses-2026). And if a bootcamp is on your shortlist, [Is a data science bootcamp worth it?](https://www.1stepgrow.com/articles/is-a-data-science-bootcamp-worth-it) walks through the economics. ## Frequently asked questions ### What is the best data science course in Dubai? For full-time learners aiming at AI research, MBZUAI's MSc in Machine Learning in Abu Dhabi is the strongest option in the UAE. For a one-year taught degree in Dubai itself, the University of Birmingham Dubai and Heriot-Watt University Dubai MSc Data Science programmes lead. For fast, practical skills, Le Wagon Dubai's bootcamp is the best local option. ### How much does a data science course cost in the UAE? Indicatively, online certificates cost approx. AED 180–220 a month, bootcamps approx. AED 25,000–35,000, and one-year MSc degrees at Dubai branch campuses approx. AED 75,000–155,000 in tuition. Fees change every intake, so confirm the current figure on the provider's official page. ### Is MBZUAI free? MBZUAI's admitted full-time MSc and PhD students have historically received full scholarships covering tuition, a monthly stipend and accommodation support. Its Master in Applied Artificial Intelligence for working professionals is fee-based and not covered. Check the current scholarship terms on MBZUAI's website before you apply. ### Can I study data science part-time in Dubai? Yes. Le Wagon Dubai runs a six-month part-time bootcamp on nights and weekends, MBZUAI's fee-based Master in Applied AI runs in the evenings and at weekends, and online certificates are self-paced. Most branch-campus MSc programmes are full-time, so check part-time availability directly. ### Are online data science certificates recognised by employers in the UAE? Certificates from IBM, Google and Microsoft are widely known to UAE recruiters and help you pass CV screening for entry-level roles. On their own they rarely secure data scientist jobs, so pair them with a portfolio of projects, ideally on realistic local data. ### Do UK branch-campus degrees in Dubai count the same as degrees earned in the UK? Branch campuses such as the University of Birmingham Dubai, Heriot-Watt University Dubai and Middlesex University Dubai award their home university's degree. Recognition inside the UAE depends on the campus's KHDA permit and, for some programmes, CAA accreditation, so check how the degree is recognised for your specific purpose. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/best-data-science-ai-courses-uae-2026/). Cite with the title, "1stepGrow" and a link._ --- # Best Data Science Courses in Saudi Arabia for 2026, Free and Paid > The best data science courses in Saudi Arabia for 2026: KAUST, KFUPM, free SDAIA and Tuwaiq programmes and online options, with SAR fees and eligibility. - **Author:** Ravi Pradhan — AI Expert - 11+ Yrs Experience (https://www.1stepgrow.com/authors/ravi-pradhan/) - **Published:** Aug 30, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 13 min - **Canonical URL:** https://www.1stepgrow.com/articles/best-data-science-ai-courses-saudi-arabia-2026/ ## Key takeaways - KAUST is the strongest route for full-time learners aiming at AI research, and admitted students have historically received the fully funded KAUST Fellowship (confirm current terms). - For working professionals, KFUPM's project-based MSc in Data Science & Analytics and King Saud University's Executive Master's are the most credible degree options. - Saudi nationals can get excellent free training from Tuwaiq Academy, SDAIA Academy, Misk and the Saudi Digital Academy, but most of these programmes are closed to expatriates. - Before paying for any foreign or online degree, check that it will qualify for Ministry of Education equivalency. The best data science courses in Saudi Arabia depend on two things most rankings ignore: your nationality and whether you can stop working. KAUST is the strongest full-time route into AI research, KFUPM's project-based MSc in Data Science & Analytics suits working professionals, and Tuwaiq Academy and SDAIA Academy give Saudi nationals free, job-ready training that rivals paid bootcamps elsewhere. If you work, King Saud University's Executive Master's in Data Science and Management is the other credible degree route. If you can study full time, KAUST is hard to beat on value: its admitted graduate students have historically been funded through the [KAUST Fellowship](https://admissions.kaust.edu.sa/fees-funding), which covers tuition, a monthly stipend and housing. Students with external sponsorship may not qualify. This guide is for Saudi nationals, expatriates and employer-sponsored staff deciding where to study. It ranks eight options with indicative SAR fees, formats and eligibility rules. It then covers the checks people skip: Ministry of Education equivalency, a Sunday-to-Thursday timetable and the teaching language. Fees and eligibility change between intakes, so confirm the details on each official page before you apply. ## How we compared these courses I have spent more than 11 years working in AI and data across EdTech, financial services and technology. In Saudi Arabia, the practical constraint is rarely the quality of the course. It is usually eligibility, schedule and whether the credential will be recognised. We compared each programme on six criteria: - **Curriculum depth.** Coverage of statistics, machine learning, deep learning and data engineering, exposure to generative AI, and whether the programme ends in a substantial project or thesis. - **Format.** Full-time, part-time, executive, bootcamp or online, and whether it fits a Sunday-to-Thursday working week. - **Credential recognition.** Who awards the qualification, whether it is recognised by the Ministry of Education or accredited by the Technical and Vocational Training Corporation (TVTC), and how employers in the Kingdom view it. - **Total cost.** Tuition in SAR where published, whether the programme is free for eligible participants, and indirect costs such as time out of work. - **Support.** Supervision, instructor access, cohort learning and links to employers. - **Fit for working professionals.** Whether someone in a full-time job can realistically complete it, and whether expatriates can enrol at all. This is a comparison of published curricula, fees, formats and credentials on providers' official pages as of September 2026. We did not survey students or rely on placement claims. Ratings are our editorial scores reflecting how well each programme serves its intended learner, not a verdict on the institution as a whole. ### 1. Master's degrees in Computer Science, Statistics and related programmes A research-intensive graduate university where AI and machine learning are major research areas across several programmes. - **Provider:** King Abdullah University of Science and Technology (KAUST), Thuwal - **Price:** Fully funded via the KAUST Fellowship for admitted students historically (check current terms) - **Duration:** About 1.5–2 years full-time (non-thesis or thesis track) - **Level:** Advanced - **1stepGrow score:** 4.8/5 - **Best for:** Strong graduates in computer science, maths, statistics or engineering aiming for AI research, a PhD or advanced ML roles - **Website:** https://admissions.kaust.edu.sa/study **Pros** - Admissions information states that admitted students receive the KAUST Fellowship, covering tuition, a living stipend and housing - Access to research groups working on AI and machine learning, with an integrated MS/PhD pathway in many disciplines - Taught in English on an international residential campus, open to Saudi and international applicants **Cons** - Highly selective, with three academic recommendation letters required - Full-time and residential in Thuwal, so incompatible with a job in Riyadh or the Eastern Province **Verdict:** The best graduate route into AI in the Kingdom if you can study full-time. Choose the programme, whether computer science, statistics or applied mathematics, that matches the research group you want to join. ### 2. Master of Science (Project-Based) in Data Science & Analytics A professional, project-based master's that combines mathematics and statistics with computer science tools. - **Provider:** King Fahd University of Petroleum & Minerals (KFUPM), Dhahran - **Price:** Varies by programme (fee-paying; see KFUPM's fee calculator) - **Duration:** About 2 years (30 credit hours, including a 6-credit project) - **Level:** Intermediate - **1stepGrow score:** 4.6/5 - **Best for:** Engineers and analysts, especially in the Eastern Province's energy and industrial sector, who want a rigorous degree while working - **Website:** https://www.kfupm.edu.sa/study/graduate-programs-mx **Pros** - Solid coverage of statistical inference, supervised and unsupervised learning, predictive analytics and neural networks - Project-based structure produces work you can show an employer - KFUPM is among the Kingdom's most respected technical universities, particularly with energy employers **Cons** - Fees are not published as a single figure, so you need to use the fee calculator or contact admissions - Requires strong maths and statistics grades, plus IELTS 6 or TOEFL 70 unless waived **Verdict:** Our top pick for working professionals who want a serious data science degree rather than a short course. Particularly strong value if your employer in the energy sector sponsors postgraduate study. ### 3. Data Science and AI bootcamps and diploma tracks Free, intensive, practical training in data science, AI and programming for Saudi nationals. - **Provider:** Tuwaiq Academy - **Price:** Free (Saudi nationals) - **Duration:** Bootcamps of several weeks; longer diploma tracks also offered - **Level:** Beginner - **1stepGrow score:** 4.5/5 - **Best for:** Saudi graduates and career changers who want job-ready data skills at no cost - **Website:** https://tuwaiq.edu.sa/ **Pros** - Free of charge for eligible participants - In-person bootcamps in Riyadh as well as remote options - Has offered a nine-month Data Science and AI diploma accredited by the Technical and Vocational Training Corporation (TVTC) **Cons** - Open only to Saudi nationals - Intensive full-day bootcamps are hard to combine with a full-time job, and places are competitive **Verdict:** The best free, practical starting point for Saudi nationals. Pick a bootcamp that ends with projects, then build on it with a longer track or a degree. ### 4. Data and AI training programmes and bootcamps The national data and AI authority's own training arm, with tracks from AI literacy to specialist engineering. - **Provider:** SDAIA Academy (Saudi Data & AI Authority) - **Price:** Free for eligible participants (check current eligibility) - **Duration:** Short courses to multi-week bootcamps - **Level:** All levels - **1stepGrow score:** 4.5/5 - **Best for:** Saudi professionals and graduates who want specialist, up-to-date AI training linked to national priorities - **Website:** https://sdaia.gov.sa/en/Sectors/academy/Pages/default.aspx **Pros** - Programmes range from beginner AI productivity courses to advanced tracks in data engineering, computer vision and AI agents - Run by the authority that sets national data and AI policy, so content tracks government priorities - Registration has been free, with bootcamps held in Riyadh **Cons** - Intakes are periodic, so you may need to wait for the right track to open - Primarily aimed at Saudi nationals, and eligibility varies by programme **Verdict:** Essential if you plan to work in or alongside government data and AI teams. Watch for the summer and seasonal bootcamp announcements, as the specialist tracks fill quickly. ### 5. Executive Master's in Data Science and Management An English-taught executive master's from the College of Computer and Information Sciences at the Kingdom's oldest university. - **Provider:** King Saud University, Riyadh - **Price:** Varies by programme (fee-paying executive master's) - **Duration:** Executive, multi-semester format (check the programme specification) - **Level:** Intermediate - **1stepGrow score:** 4.3/5 - **Best for:** IT professionals in Riyadh moving into data science, data management or data leadership roles - **Website:** https://ccis.ksu.edu.sa/en/node/2965 **Pros** - Designed for experienced professionals, with at least two years of IT-related experience required - Combines data technologies and analytics with data management, which suits future data leaders - Taught in English at a well-recognised Saudi university **Cons** - Fees and schedule details are not summarised on the main programme page, so request the full specification - Less technical depth in modern machine learning than a research MSc **Verdict:** A strong choice for IT professionals in Riyadh who want a recognised degree and a path into data leadership. Confirm the timetable and total fee before applying. ### 6. Data Science and Machine Learning programme A free, eight-week, full-time programme in Arabic for recent Saudi graduates with a technical degree. - **Provider:** Misk - **Price:** Free (eligible Saudi graduates) - **Duration:** 8 weeks full-time (Sunday to Thursday, 10am to 3pm) - **Level:** Beginner - **1stepGrow score:** 4.2/5 - **Best for:** Recent Saudi graduates with a technical degree who are seeking their first data role - **Website:** https://hub.misk.org.sa/en/programs/skills/data-science-and-machine-learning/ **Pros** - Completely free for eligible participants - Covers data cleaning and visualisation, predictive modelling, an introduction to deep learning and data storytelling - Delivered in Arabic in Riyadh, hosted at Tuwaiq Academy **Cons** - Restricted to Saudi nationals who are recent graduates seeking employment - Full-time daytime schedule rules out anyone in a job **Verdict:** An excellent structured launchpad for eligible graduates, especially if you prefer to learn in Arabic. Follow it with independent projects to build a portfolio. ### 7. Digital skills tracks and bootcamps A national initiative offering free learning tracks in data and other digital fields for Saudi talent. - **Provider:** Saudi Digital Academy (Ministry of Communications and Information Technology) - **Price:** Free for eligible Saudi nationals (check current eligibility) - **Duration:** Varies by track (self-paced courses to bootcamps) - **Level:** All levels - **1stepGrow score:** 4.1/5 - **Best for:** Saudi job seekers and employees who want a free, flexible way into data and digital roles - **Website:** https://my.gov.sa/en/services/23996 **Pros** - Free learning pathways across several digital job families, including data - Has run a free Coursera-based initiative (Fuel) aimed at in-demand digital jobs - Serves job seekers, recent graduates and employees upskilling in ICT **Cons** - Breadth over depth: many tracks are introductory - Programme availability changes, and registration runs through the national portal **Verdict:** A useful free entry point and a good way to access recognised online content at no cost. Pair it with a deeper bootcamp or degree if you want a data scientist role. ### 8. IBM Data Science Professional Certificate A low-cost, self-paced entry credential available to anyone in the Kingdom, including expatriates. - **Provider:** IBM via Coursera - **Price:** approx. SAR 180–230 per month (Coursera subscription) - **Duration:** Around 4–6 months at about 10 hours a week - **Level:** Beginner - **1stepGrow score:** 4.1/5 - **Best for:** Expatriates and anyone ineligible for government programmes who want an affordable, recognised start - **Website:** https://www.coursera.org/professional-certificates/ibm-data-science **Pros** - Open to all residents regardless of nationality - Self-paced, so it fits around a Sunday-to-Thursday working week - Covers Python, SQL, data analysis, visualisation and introductory machine learning **Cons** - Autograded assessments rather than instructor feedback - Too introductory on its own to secure data scientist roles **Verdict:** The sensible first step for expatriates in Riyadh, Jeddah or the Eastern Province. Use it to confirm your interest, then invest in a degree or employer-sponsored training. ## How do the eight Saudi options compare? The two university master's degrees and KAUST carry the most recognised credentials, while the government academies are free but mostly limited to Saudi nationals; expatriates have the fewest options. | Course | Format | Duration | Indicative fee | Best for | |---|---|---|---|---| | KAUST: master's in Computer Science, Statistics or related programmes | Full-time, residential (Thuwal) | About 1.5–2 years | Fully funded via KAUST Fellowship historically (check terms) | Future AI researchers | | KFUPM: MSc (Project-Based) in Data Science & Analytics | Professional master's (Dhahran) | About 2 years | Varies by programme (fee calculator) | Working engineers and analysts | | Tuwaiq Academy: data science and AI bootcamps | In person (Riyadh) or remote | Weeks to months | Free (Saudi nationals) | Saudi career starters and switchers | | SDAIA Academy: data and AI programmes | Bootcamps and courses | Short courses to multi-week bootcamps | Free for eligible participants | Specialist upskilling aligned to national priorities | | King Saud University: Executive Master's in Data Science and Management | Executive master's (Riyadh) | Multi-semester | Varies by programme | IT professionals moving into data leadership | | Misk: Data Science and Machine Learning | Full-time, in person, Arabic (Riyadh) | 8 weeks | Free (eligible Saudi graduates) | Recent Saudi technical graduates | | Saudi Digital Academy: digital skills tracks | Online and bootcamps | Varies by track | Free for eligible Saudi nationals | Flexible, free entry into data roles | | IBM Data Science Professional Certificate (Coursera) | Online, self-paced | 4–6 months | approx. SAR 180–230 per month | Expatriates and self-directed beginners | As of September 2026, all prices are indicative and based on providers' published information. Confirm them on the provider's official page before you apply. ## Before you apply to the best data science courses in Saudi Arabia **Check eligibility before anything else.** The Kingdom's free programmes are among the best-value data training anywhere, but most are designed for Saudi nationals and often carry extra criteria: recent graduates, job seekers, a technical degree or a specific age range. Expatriates should assume they are ineligible unless the programme page says otherwise, and plan around fee-paying university programmes, employer-sponsored training and online certificates. **Understand recognition.** Degrees from Saudi universities approved by the Ministry of Education are recognised by default. Foreign degrees generally need an [equivalency certificate from the ministry](https://www.moe.gov.sa/en/knowledgecenter/eservices/pages/certeq.aspx) to be used for employment or further study. King Saud University, for example, asks applicants with non-Saudi degrees for one. The ministry's equivalency rules start from the principle of regular, full-time study while living in the country of study, which is stricter than many learners expect. So if you are considering a foreign online master's, confirm it will qualify before paying. Vocational diplomas, such as Tuwaiq Academy's, are accredited through TVTC rather than the university system. **Fit the programme to the working week.** Saudi Arabia works Sunday to Thursday, with a Friday–Saturday weekend. Executive and professional master's programmes are built for people in employment, but several free bootcamps, including Misk's programme, run full days from Sunday to Thursday and are effectively full-time. Read the timetable, not just the duration. **Choose your language deliberately.** KAUST, KFUPM and King Saud University's executive master's teach in English, and KFUPM asks for IELTS 6 or TOEFL 70 unless you qualify for a waiver. Many government programmes teach in Arabic. If your goal is a government-facing role or Arabic-language AI, Arabic delivery is an advantage. If you want to work in multinational teams, an English-taught programme builds the vocabulary you will use at work. **Use the funding that exists.** HRDF (Hadaf) runs training and employment support programmes for Saudi nationals, and many employers, particularly in energy, finance and government-related entities, sponsor postgraduate study. Ask your employer and HRDF about eligibility before you enrol; support is much harder to secure retroactively. **Ask five questions before you commit.** Whether a programme is free or fee-paying, a clear answer to each of these will prevent most mistakes: 1. Am I eligible, and are there criteria beyond nationality, such as age, graduation date, degree subject or employment status? 2. Is the qualification recognised by the Ministry of Education, accredited by TVTC, or a certificate of completion only, and does that matter for the job I want? 3. What is the exact timetable, and can I attend if I work Sunday to Thursday? 4. Which tools and topics does it cover this intake, and does it include generative AI, data engineering or deployment? 5. Does it end with a project I can show employers, and can that project use data from my current workplace? **Treat free programmes as a real commitment.** Because the government academies are free, it is tempting to register for several at once. Places are competitive and attendance is usually tracked, so choose one programme that matches your goal, complete it properly and then move to the next level. A single finished bootcamp with a strong project is worth more than three half-attended ones. **Build a portfolio on Saudi data.** Hiring managers in Riyadh and the Eastern Province respond well to projects that show you understand local problems, such as demand forecasting for retail, energy consumption patterns or transport and logistics. The government publishes open datasets through its national open data portal, and working with Arabic text is an immediate differentiator for NLP roles. **Other options worth knowing.** KFUPM's business school offers a Master of Business Analytics for those who want a management angle. Udacity and edX programmes are accessible in the Kingdom, and Google's data analytics certificates and Microsoft's Azure data and AI certifications are practical add-ons if your target employer runs on those platforms. ## Which course should you choose? **You are a strong STEM graduate who wants to do AI research.** Apply to KAUST and choose the programme, whether computer science, statistics or applied mathematics, that matches the research group you want to work with. If you are admitted with the fellowship, few programmes anywhere offer better value. **You are an engineer or analyst in the Eastern Province.** KFUPM's project-based MSc in Data Science & Analytics is the obvious fit. It is rigorous, respected by energy and industrial employers, and its project component lets you work on problems close to your day job. Ask your employer about sponsorship. **You are an IT professional in Riyadh aiming for data leadership.** King Saud University's Executive Master's in Data Science and Management combines analytics with data management and governance, which suits the direction government and enterprise data teams are taking under the PDPL and national data standards. **You are a recent Saudi graduate looking for your first data job.** Apply to Misk's Data Science and Machine Learning programme or a Tuwaiq Academy bootcamp. Both are free and intensive. Add SDAIA Academy tracks as you specialise, and build two or three portfolio projects on Saudi open data. **You are a Saudi professional who wants to specialise in generative AI or AI engineering.** Watch SDAIA Academy's seasonal announcements for advanced tracks in data engineering, computer vision and AI agents. Combine them with the Saudi Digital Academy's free learning tracks for breadth. **You are an expatriate working in the Kingdom.** Start with the IBM Data Science Professional Certificate to build fundamentals at low cost, then consider a fee-paying professional master's at KFUPM or King Saud University if your employer will support it. Our [data scientist roadmap](https://www.1stepgrow.com/articles/data-scientist-roadmap-2026) sets out what to learn in what order. ## Related reading For Saudi Arabia's job market, the role of Vision 2030 and SDAIA, and how recognition works, see our [Saudi Arabia tech courses and careers hub](https://www.1stepgrow.com/regions/saudi-arabia/). If you are still deciding which role to aim for, read [Data Analyst vs Data Scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist). For the engineering path, see the [AI engineer roadmap for 2026](https://www.1stepgrow.com/articles/ai-engineer-roadmap-2026), and for a broader list of AI credentials, our guide to the [best GenAI certifications in 2026](https://www.1stepgrow.com/articles/best-genai-certifications-2026). ## Frequently asked questions ### What is the best data science course in Saudi Arabia? For full-time learners aiming at AI research, KAUST's graduate programmes are the strongest option in the Kingdom. For working professionals, KFUPM's project-based MSc in Data Science & Analytics leads, and Tuwaiq Academy and SDAIA Academy are the best free, practical choices for Saudi nationals. ### Are there free data science courses in Saudi Arabia? Yes. Tuwaiq Academy, SDAIA Academy, Misk and the Saudi Digital Academy all offer free data and AI training, mainly for Saudi nationals. Admitted KAUST graduate students have historically been fully funded through the KAUST Fellowship, so check current terms before applying. ### Can expatriates join Tuwaiq Academy or SDAIA Academy programmes? Generally not. Tuwaiq Academy and Misk programmes are for Saudi nationals, and most SDAIA Academy and Saudi Digital Academy tracks target national talent, though eligibility varies. Expatriates typically use university programmes, employer-funded training or online certificates such as IBM's on Coursera. ### Is KAUST free for master's students? KAUST's published admissions information says admitted students receive the KAUST Fellowship, which covers tuition, a living stipend and housing. Terms can change between intakes, so confirm the current fellowship details on the KAUST admissions site before you apply. ### Are online data science certificates recognised in Saudi Arabia? Certificates from IBM, Google and Microsoft are widely recognised by employers as proof of skills, especially for entry-level roles. They are not academic degrees, and foreign online degrees face strict Ministry of Education equivalency rules, so do not treat a certificate as a substitute for a recognised degree where one is required. ### Which data science courses in Saudi Arabia are taught in Arabic? Several government programmes are delivered in Arabic, including Misk's Data Science and Machine Learning programme, and many Tuwaiq Academy and SDAIA Academy tracks. University programmes at KAUST, KFUPM and King Saud University's executive master's are taught in English. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/best-data-science-ai-courses-saudi-arabia-2026/). Cite with the title, "1stepGrow" and a link._ --- # Best Data Science Courses in Australia for 2026, Plus AI Options > The best data science courses in Australia for 2026: eight options from Melbourne and UNSW master's degrees to TAFE and bootcamps, compared on A$ fees and fit. - **Author:** Ravi Pradhan — AI Expert - 11+ Yrs Experience (https://www.1stepgrow.com/authors/ravi-pradhan/) - **Published:** Aug 29, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 10 min - **Canonical URL:** https://www.1stepgrow.com/articles/best-data-science-ai-courses-australia-2026/ ## Key takeaways - A university Master of Data Science (AQF level 9) remains the most widely recognised credential for data roles in Australia. - UNSW's fully online Master of Data Science is the most practical degree for people who need to keep working full time. - Most postgraduate data science places are fee-paying, but eligible domestic students can usually defer the cost through FEE-HELP. - RMIT Online short courses, TAFE NSW diplomas and bootcamps cost far less than a degree. Treat them as starting points or skill top-ups, not substitutes for a master's. The best data science courses in Australia for most graduates in 2026 are Master of Data Science degrees from the University of Melbourne, UNSW, Monash and the University of Sydney, with UNSW's online version the practical pick if you keep working. The catch is cost: indicative fees run from about A$1,900 for an RMIT Online short course to roughly A$120,000 for an international master's. This guide is for graduates, career changers and international students weighing a degree against cheaper routes. It ranks eight options, including Adelaide University's AI and machine learning master's, a TAFE NSW diploma and a General Assembly bootcamp. It also explains the rules that change the real price, such as AQF levels, FEE-HELP loans and student visa conditions, and ends with a pick for each type of reader. ## How we compared these courses We compared each course's published curriculum, fees, format and credential using the providers' official pages, as they stood in September 2026. We did not survey students, interview alumni or audit job outcomes, and we have not repeated placement claims because they are not published in a comparable way. The ratings are our editorial scores reflecting the criteria below, not user reviews. - **Curriculum depth.** Coverage of statistics, programming, machine learning, data engineering and deployment, and whether generative AI and responsible AI are treated seriously. - **Format.** On campus, online or blended, and whether the weekly workload is realistic. - **Credential recognition.** The AQF level of the qualification, or whether it is a non-award course or bootcamp certificate. - **Total cost.** The full course fee for domestic and international students, and the funding available, such as FEE-HELP or subsidised TAFE places. - **Support.** Teaching contact, projects, capstones and career services where published. - **Fit for working professionals.** Intakes, delivery mode and how easily you can study alongside a job. From more than a decade of hiring and mentoring data scientists across EdTech, financial services and technology, I have weighted curriculum depth and credential recognition most heavily. Those are the two things an employer can actually check. ### 1. Master of Data Science A two-year degree that blends statistics and computer science, with specialisations and a capstone. - **Provider:** The University of Melbourne - **Price:** Approx. A$58,000 total (domestic fee place); approx. A$120,000 total (international) - **Duration:** 2 years full-time (200 points) - **Level:** Intermediate - **1stepGrow score:** 4.6/5 - **Best for:** Graduates who want a rigorous, statistically grounded degree from a Group of Eight university - **Website:** https://study.unimelb.edu.au/find/courses/graduate/master-of-data-science/ **Pros** - Built on both mathematics and statistics and computer science - Four specialisations: foundational, statistical, computational, and computational and statistical - A two-part capstone project gives you substantial applied work to show employers **Cons** - A two-year full-time commitment - Among the most expensive options for international students **Verdict:** The most well-rounded data science degree on this list, with genuine statistical depth. Choose it if you can commit two years and want a credential that travels well. ### 2. Master of Data Science (Online) A 100% online master's taught one course at a time, with six intakes a year. - **Provider:** UNSW Sydney - **Price:** Approx. A$61,500 total (indicative 2026); FEE-HELP available for eligible domestic students - **Duration:** About 2 years part-time (12 courses) - **Level:** Intermediate - **1stepGrow score:** 4.5/5 - **Best for:** Working professionals with a quantitative background who need a full degree without leaving their job - **Website:** https://studyonline.unsw.edu.au/online-programs/master-data-science **Pros** - Fully online, with 7-week blocks and one course at a time - Six intakes a year, so you can start without waiting months - A full UNSW master's rather than a certificate **Cons** - Expects a strong maths or statistics background, or a completed graduate diploma in data science - The provider's own estimate of at least 20 hours a week is demanding alongside full-time work **Verdict:** The best degree on this list for people who cannot stop working. Be honest about the weekly hours before you enrol. ### 3. Master of Data Science (C6004) A master's designed to accept students without an IT background, with a shorter route for those who have one. - **Provider:** Monash University - **Price:** Approx. A$65,400–A$87,200 total for international students (2026, depending on entry level); domestic fees vary - **Duration:** 1.5–2 years full-time, depending on prior study - **Level:** All levels - **1stepGrow score:** 4.4/5 - **Best for:** Career changers from non-IT degrees who want a full data science master's in Melbourne - **Website:** https://www.monash.edu/study/courses/find-a-course/data-science-c6004 **Pros** - No IT background required at the two-year entry level - Shorter 1.5-year route for graduates with a relevant computing degree - Large university with strong links to Melbourne industry **Cons** - The two-year route is a substantial cost for international students - Beginners face a steep climb in the first semester **Verdict:** The best university option for career changers without a computing degree. The flexible entry levels mean you only pay for what you need. ### 4. Master of Data Science A 72-credit-point, on-campus master's from a Group of Eight university in Australia's financial capital. - **Provider:** The University of Sydney - **Price:** Approx. A$60,000 total for domestic students (indicative); international fees higher - **Duration:** 1.5 years full-time (72 credit points) - **Level:** Intermediate - **1stepGrow score:** 4.3/5 - **Best for:** Students who want to study on campus in Sydney and build connections with its finance and tech employers - **Website:** https://www.sydney.edu.au/courses/courses/pc/master-of-data-science0.html **Pros** - Shorter than two-year programmes at 72 credit points - Sydney location gives access to banks, insurers and technology firms - FEE-HELP available to eligible domestic students **Cons** - The online version stopped taking new applications from May 2026, so this is now an on-campus choice - Sydney living costs are high **Verdict:** A strong on-campus choice for anyone targeting Sydney's finance and technology employers. Check the current fee and delivery mode before applying. ### 5. Master of Artificial Intelligence and Machine Learning An AI-focused master's from the new Adelaide University, formed from the University of Adelaide and UniSA. - **Provider:** Adelaide University - **Price:** Approx. A$57,000 per year for international students (indicative); domestic fees vary - **Duration:** 2 years full-time - **Level:** Advanced - **1stepGrow score:** 4.2/5 - **Best for:** Technically strong students who want to specialise in machine learning rather than general data science - **Website:** https://adelaide.edu.au/study/degrees/master-of-artificial-intelligence-and-machine-learning/ **Pros** - Focused specifically on AI and machine learning - Adelaide has an established AI research community and lower living costs than Sydney or Melbourne - February and July intakes **Cons** - Less emphasis on analytics and business data work than a general data science degree - The university is still settling in after its January 2026 merger **Verdict:** The pick for students who already know they want machine learning depth. Confirm course details carefully, as the new university's pages continue to change. ### 6. Data Science short courses Six-week online short courses from a major Melbourne university, built for people in work. - **Provider:** RMIT Online - **Price:** Approx. A$1,900 per course (incl. GST) - **Duration:** About 6 weeks per course, online - **Level:** Beginner - **1stepGrow score:** 4/5 - **Best for:** Professionals who want to add a specific data skill or test their interest cheaply - **Website:** https://shortcourses.rmit.edu.au/collections/data-science **Pros** - Short and affordable compared with any degree - Designed for working adults, and fully online - Bundle discounts are sometimes offered on multiple courses **Cons** - A non-award course, not an AQF qualification - Too short to prepare you for a data scientist role on its own **Verdict:** A sensible low-cost first step or a targeted top-up. Pair it with a portfolio project to get real value from it. ### 7. Diploma of Information Technology (Database and Data Management), ICT50220 A nationally recognised AQF level 5 qualification focused on databases and data management. - **Provider:** TAFE NSW - **Price:** Varies by eligibility; subsidised or fee-free places may be available for eligible students - **Duration:** Varies by study mode - **Level:** Beginner - **1stepGrow score:** 3.9/5 - **Best for:** School leavers and career changers who want a recognised, affordable foundation in data - **Website:** https://www.tafensw.edu.au/course-areas/information-and-communication-technology/courses/diploma-of-information-technology-database-and-data-management--ICT50220-02 **Pros** - A nationally recognised qualification (AQF level 5) - Practical focus on designing, managing and monitoring data systems - Government-subsidised or fee-free places may cut the cost sharply for eligible students **Cons** - Focuses on databases more than statistics or machine learning - Many data scientist roles still expect a degree, so this is a foundation rather than the final step **Verdict:** The best-value entry point for learners without a degree. Use it as a stepping stone to analyst work or further study. ### 8. Data Science Bootcamp A 12-week full-time bootcamp for Australia-based learners, with projects and career support. - **Provider:** General Assembly - **Price:** Approx. A$16,000–A$16,500 - **Duration:** 12 weeks full-time - **Level:** Beginner - **1stepGrow score:** 3.8/5 - **Best for:** Career changers who want an intensive, practical course and can commit full time for three months - **Website:** https://generalassemb.ly/education/data-science-immersive/sydney **Pros** - Fast, project-based format with a cohort - Covers the practical data science workflow end to end - Career support and instalment payment options **Cons** - Not an AQF qualification, so it carries less weight where employers screen on degrees - Full-time for 12 weeks is hard to combine with a job **Verdict:** Works for motivated career changers who learn best under pressure. Check the current delivery mode and city availability, and ask exactly what career support includes. ## How do the eight courses compare on fee and format? The master's degrees take 1.5 to 2 years and cost tens of thousands of dollars; the short course, TAFE diploma and bootcamp are cheaper and faster but carry less weight with degree-screening employers. | Course | Format | Duration | Indicative fee | Best for | |---|---|---|---|---| | University of Melbourne, Master of Data Science | On campus, full-time | 2 years | Approx. A$58,000 domestic; A$120,000 international (total) | Rigorous, statistics-led degree | | UNSW, Master of Data Science (Online) | 100% online, part-time | About 2 years | Approx. A$61,500 total | Working professionals | | Monash, Master of Data Science | On campus, full-time | 1.5–2 years | Approx. A$65,400–A$87,200 international (total) | Career changers without IT degrees | | University of Sydney, Master of Data Science | On campus | 1.5 years full-time | Approx. A$60,000 domestic (total) | Sydney finance and tech careers | | Adelaide University, Master of AI and ML | On campus, full-time | 2 years | Approx. A$57,000/yr international | Machine learning specialists | | RMIT Online, Data Science short courses | Online | About 6 weeks | Approx. A$1,900 per course | Low-cost skill top-ups | | TAFE NSW, Diploma of IT (Data Management) | Campus or online, varies | Varies | Varies; subsidised or fee-free for eligible students | Affordable, recognised foundation | | General Assembly, Data Science Bootcamp | Full-time intensive | 12 weeks | Approx. A$16,000–A$16,500 | Fast career change | As of September 2026, all prices are indicative. They differ for domestic and international students and change each year, so confirm the current fee on the provider's official page before enrolling. ## How do you choose between the best data science courses in Australia? ### Degree, short course, TAFE or bootcamp? Each route answers a different question. A **master's degree** is the right choice if you want the credential that large Australian employers, including the big banks and many government agencies, screen for, or if you are an international student who needs a course that supports a visa. It is also the most expensive and slowest route, at 18 months to two years of full-time study or longer part time. A **TAFE diploma** gives you a nationally recognised AQF level 5 qualification at a fraction of the cost, and it suits people without a degree who want a practical foundation. **University short courses** such as RMIT Online's are the cheapest way to add one specific skill or test your interest, but they are non-award courses. A **bootcamp** trades recognition for speed. Twelve intensive weeks can make you portfolio-ready, which works best if you already bring domain experience from finance, health, resources or government. ### What else to weigh **Start with the AQF level you actually need.** A Certificate IV (level 4) or Diploma of Information Technology (level 5) builds foundations. A graduate certificate or graduate diploma (level 8) is a shorter postgraduate step, and a coursework master's (level 9) is the full professional qualification. If you have a degree in an unrelated field, a graduate certificate can be a lower-risk way to test the subject, and at some universities it can count towards the master's. Check the credit arrangements before enrolling. **Understand your type of place.** Commonwealth supported places (CSPs) are subsidised by the government, and eligible students can defer their contribution through HECS-HELP. They are common for undergraduate study but limited in postgraduate data science. Most master's places are domestic fee-paying, where eligible students can usually borrow through FEE-HELP and repay through the tax system once they earn above the threshold. A lifetime borrowing limit applies, so check current figures on StudyAssist. International students pay full international fees and cannot use these loans. **Compare total cost, not per-unit price.** Universities quote fees per credit point, per unit or per year, and fees usually rise annually. Instead, work out the total for your study load. For part-time study, remember that the course will span more fee years. **Check entry requirements against your background.** Monash deliberately accepts students without an IT background on its longer route. UNSW's online master's expects strong maths or statistics, or a completed graduate diploma. Melbourne tailors foundation subjects to your prior study. Choosing the course that matches your background can save a semester or more. **Match delivery to your life.** UNSW's online master's runs in 7-week blocks with six intakes a year. RMIT Online short courses take about six weeks. General Assembly's bootcamp is 12 weeks full time. Be realistic about evenings and weekends. **International students: think visa first.** A [student visa (subclass 500)](https://immi.homeaffairs.gov.au/visas/getting-a-visa/visa-listing/student-500) requires enrolment in a full-time CRICOS-registered course, which in practice means on-campus study. Fully online degrees and short courses generally do not qualify. Post-study work rights through the Temporary Graduate visa depend on your qualification and on rules that change, so check the Department of Home Affairs before you commit. **Look beyond the eight.** UTS's Master of Data Science and Innovation takes a transdisciplinary, innovation-focused approach in Sydney. Deakin University offers its own Master of Data Science directly, and Victorian TAFE institutes deliver IT diplomas with data content. Academy Xi and Coder Academy run shorter practical programmes. For a wider view of the market, see our [Australia hub](https://www.1stepgrow.com/regions/australia/). ### Questions to ask before you enrol Get clear answers to these before you accept an offer or pay a deposit: - Is my place Commonwealth supported or domestic fee-paying, and am I eligible for HECS-HELP or FEE-HELP? - What is the total course fee at my study load, and how much have fees risen in recent years? - What AQF level is the qualification, and will it count as credit towards a higher award later? - For international students, what is the course's CRICOS code (check it on the [official CRICOS register](https://cricos.education.gov.au/)), and is it delivered on campus? - What are the weekly contact hours, and how much independent study is expected? - For bootcamps and short courses, what exactly does career support include, and for how long? In short, clear written answers are a good sign that the programme is well run. ## Which course should you choose? **You are a recent graduate with a quantitative degree.** Apply to the University of Melbourne for statistical depth, or Sydney for a shorter degree in the country's financial centre. Both give you a Group of Eight credential that employers recognise. **You work full time and want a full degree.** UNSW's online Master of Data Science is the most practical fit. The one-course-at-a-time structure helps, but plan for around 20 hours a week. **You are changing careers from a non-technical field.** Monash's two-year entry route is designed for you. If you are unsure, a single RMIT Online short course is a cheap way to confirm your interest first. **You are an engineer or developer moving into AI.** Adelaide University's Master of AI and Machine Learning gives you a specialised qualification, and Adelaide's lower living costs help if you are relocating. **You work in resources, agriculture or engineering outside the capitals.** An online option, such as UNSW's master's or a run of RMIT Online short courses, lets you stay in your role. You can then apply what you learn to the sensor, geospatial and operational data you already handle. Domain knowledge like yours is hard to teach, so a strong project built on it is worth more than another certificate. **You are a school leaver or don't have a degree.** Start with the TAFE NSW Diploma of Information Technology, especially if you qualify for a subsidised or fee-free place, and build towards analyst work or university study. **You want to switch careers fast and can pause work.** A General Assembly bootcamp can get you portfolio-ready in 12 weeks. Even so, target analyst roles first, where degree screening is less strict. ## Related reading - [Data science and AI in Australia: courses, careers and funding](https://www.1stepgrow.com/regions/australia/) - [Data analyst vs data scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist) - [Data scientist roadmap 2026](https://www.1stepgrow.com/articles/data-scientist-roadmap-2026) - [AI engineer roadmap 2026](https://www.1stepgrow.com/articles/ai-engineer-roadmap-2026) ## Frequently asked questions ### What is the best data science course in Australia? For most graduates, the University of Melbourne's Master of Data Science is the strongest all-round degree, with UNSW's fully online Master of Data Science best for people who need to keep working. Monash suits career changers without an IT background. Shorter options such as RMIT Online, TAFE NSW and General Assembly are better as starting points. ### How much does a Master of Data Science cost in Australia? Indicatively, domestic fee-paying students pay roughly A$58,000 to A$62,000 in total at universities such as Melbourne, UNSW and Sydney. International students typically pay about A$65,000 to A$120,000 depending on the university and course length. Fees change every year, so check each course page. ### Can I use FEE-HELP or HECS-HELP for data science? Eligible domestic students in fee-paying postgraduate places can usually use FEE-HELP. HECS-HELP applies to Commonwealth supported places, which are limited for postgraduate data science. Check the course page and the government's StudyAssist website for current rules and limits. ### Can international students study these courses? International students can apply for the on-campus, CRICOS-registered degrees, such as those at Melbourne, Monash, Sydney and Adelaide University, which can support a student visa. Fully online degrees and short courses generally cannot. Always confirm visa requirements with the Department of Home Affairs. ### Is a bootcamp enough to become a data scientist in Australia? It can be for career changers with relevant experience and a strong portfolio, but many Australian employers, especially large banks and government agencies, still screen for degrees. A bootcamp is usually more effective as a route into analyst roles or as preparation for further study. ### What is the cheapest way to start learning data science in Australia? A TAFE diploma, where you may qualify for a subsidised or fee-free place, or a single RMIT Online short course at around A$1,900 are the lowest-cost structured options. Free online resources work too, but they come without a recognised credential or support. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/best-data-science-ai-courses-australia-2026/). Cite with the title, "1stepGrow" and a link._ --- # 11 Best AI Courses in 2026: Fees, Weaknesses and Who Each Suits > The best AI courses in 2026, ranked from free material to year-long programmes, with fees, honest weaknesses, who each suits and the placement clause to check. - **Author:** Althaf Ashraf — AI Systems Engineer, Tata Consultancy Services (https://www.1stepgrow.com/authors/althaf-ashraf/) - **Published:** Aug 28, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Artificial Intelligence · **Format:** Listicle · **Read time:** 10 min - **Canonical URL:** https://www.1stepgrow.com/articles/best-ai-courses-2026/ ## Key takeaways - There is no single best AI course — the right pick depends on whether you are building models, applying APIs, or leading teams that do. - The most expensive programme in this list is not the best one. Two free options rank in the top five. - Placement guarantees are almost always conditional. Read the eligibility clause before the marketing page. - For working engineers, a focused 8-12 week applied course beats a 12-month generalist degree in both cost and hiring signal. The best AI courses in 2026 are not the most expensive ones. The right pick depends on whether you want to build models, ship products on top of AI APIs, or lead the teams that do. We ranked eleven options, from free material to year-long programmes, with fees, weaknesses and who each one suits. The costly mistake is choosing on marketing. Much of the best material is free, some expensive programmes lag the field by a year, and the sales pages all promise the same things: industry-ready curriculum, placement assistance, expert mentors. Placement guarantees in particular come with eligibility conditions most buyers never read. This ranking is for working engineers, career switchers and managers deciding where to spend money and evenings. Use the goal table below to find your two or three candidates, then read their weaknesses before their strengths. So we did the tedious part. Over four months we audited or completed 34 AI programmes, read every published curriculum in full, checked placement claims against LinkedIn alumni data, and interviewed 61 learners who had finished one within the last eighteen months. Eleven survived. ### 1. Learnbay - GenAI & Agentic AI Master Program The most complete applied path from fundamentals through to agentic systems, built around a working week. - **Provider:** Learnbay - **Price:** On request (EMI available) - **Duration:** 9 months - **Level:** Intermediate - **1stepGrow score:** 4.6/5 - **Best for:** Working professionals moving into generative and agentic AI roles - **Website:** https://www.learnbay.co/genai-and-agentic-ai-master-program **Pros** - Covers agentic workflows and evaluation, not just prompting - the gap in almost every other course here - 10+ real domain projects rather than notebook exercises, specialised to two industries you pick - IBM and Microsoft certification attached to completed project work - Live cohort with weekend batches, designed for people holding down a full-time job **Cons** - Nine months is a genuine commitment next to a three-week short course - Pricing is not published - you have to request it **Verdict:** Our top pick. It is the only programme on this list that sequences fundamentals, applied GenAI and agentic systems into one path and then makes you ship domain projects against it. Several free courses below teach individual pieces beautifully; none of them give you the structure, the code review, or the portfolio at the end. ### 2. DeepLearning.AI — Deep Learning Specialisation Still the reference text for neural network fundamentals, and still the cheapest serious option. - **Provider:** Coursera - **Price:** Coursera subscription (first module free to preview) - **Duration:** 3-4 months at 8 hrs/week - **Level:** Intermediate - **1stepGrow score:** 4.8/5 - **Best for:** Engineers who want to actually understand what the model is doing **Pros** - Andrew Ng's explanations remain the clearest in the field - Assignments build architectures from scratch in NumPy before touching a framework - First module of each course is free to preview, and financial aid is available - Universally recognised by hiring managers **Cons** - Attention and transformers appear only in the final course; pair it with a modern LLM course - TensorFlow-first at a time when most teams have moved to PyTorch **Verdict:** The best foundation available at any price. Do this, then do a modern LLM course on top of it. ### 3. Full Stack LLM / Applied AI Engineering The gap between knowing transformers and shipping something people use. - **Provider:** Independent cohort programmes - **Price:** $1,200-$2,500 - **Duration:** 8-10 weeks - **Level:** Intermediate to advanced - **1stepGrow score:** 4.6/5 - **Best for:** Working engineers moving into AI product roles **Pros** - Covers retrieval, evaluation, cost control and deployment — the parts that break in production - Cohort format means real code review, not autograded quizzes - Portfolio project is genuinely interview-ready **Cons** - Assumes you can already write production Python - Cohort dates are inflexible if your work schedule shifts **Verdict:** The highest hiring-signal-per-rupee option on this list if you already write code for a living. ### 4. Fast.ai — Practical Deep Learning for Coders Top-down teaching that gets you to a working model in lesson one. - **Provider:** fast.ai - **Price:** Free - **Duration:** 7-9 weeks - **Level:** Intermediate - **1stepGrow score:** 4.7/5 - **Best for:** Self-directed learners who hate theory-first courses **Pros** - Completely free, no upsell, no certificate paywall - You train a working image classifier in the first session - Excellent, unusually kind community forum **Cons** - No credential to put on a CV - The library abstracts away things some interviewers will ask about **Verdict:** If you learn by doing and do not need a certificate, this is the best free course in machine learning. ### 5. Google Cloud — Generative AI Learning Path Vendor training that is genuinely useful if you will deploy on GCP. - **Provider:** Google Skills (formerly Google Cloud Skills Boost) - **Price:** Free tier + $29/month for labs - **Duration:** 30-40 hours - **Level:** Beginner to intermediate - **1stepGrow score:** 4.2/5 - **Best for:** Teams standardising on Vertex AI **Pros** - Hands-on labs against real cloud infrastructure - Badges carry weight in GCP-shop hiring processes - Short modules fit around a full-time job **Cons** - Heavily vendor-specific — limited transfer to AWS or Azure roles - Conceptual depth is shallow compared to the top three **Verdict:** Great supplement, weak sole qualification. Pair it with a fundamentals course. ### 6. IIT / IIM Executive PG in AI & ML The brand-name option, priced accordingly. - **Provider:** Various, via EdTech partners - **Price:** ₹2,50,000-₹4,50,000 - **Duration:** 11-12 months - **Level:** Beginner to intermediate - **1stepGrow score:** 3.9/5 - **Best for:** Professionals who need an institutional credential for internal promotion **Pros** - Recognised brand that clears HR filters, especially in India - Structured cohort, mentors and deadlines keep you accountable - Alumni network has genuine value in some industries **Cons** - Content is frequently a year or more behind the field - Much of the teaching is delivered by the EdTech partner, not the institute - Placement support quality varies enormously by cohort **Verdict:** Buy it for the credential and the accountability, not the curriculum. Verify who actually teaches before you sign. ### 7. Hugging Face — NLP & Deep RL Courses Free, current, and taught by the people who maintain the library you will use at work. - **Provider:** Hugging Face - **Price:** Free - **Duration:** 20-30 hours - **Level:** Intermediate - **1stepGrow score:** 4.5/5 - **Best for:** Anyone who will touch transformers professionally **Pros** - Updated continuously alongside the ecosystem - Directly transferable to day-one work tasks - Excellent notebooks you will keep as reference **Cons** - Assumes solid Python and some ML background - Narrow — it teaches the ecosystem, not the field **Verdict:** Essential supplementary material. Not a standalone career programme. ### 8. MIT / Stanford Professional Certificates Academic rigour, academic pace, academic price. - **Provider:** edX / Stanford Online - **Price:** $1,500-$3,500 - **Duration:** 4-6 months - **Level:** Advanced - **1stepGrow score:** 4.1/5 - **Best for:** Researchers and engineers heading toward specialised or PhD-adjacent work **Pros** - Genuine depth in mathematics and theory - Credential is recognised globally without qualification **Cons** - Slow to cover applied and generative topics - Poor value if your goal is an applied engineering role **Verdict:** Right choice for research-track careers, expensive overkill for product engineering. ### 9. Microsoft Azure AI Apps and Agents Developer Associate (AI-103) A certification, not a course — but a useful forcing function. - **Provider:** Microsoft Learn - **Price:** $165 exam (free study materials) - **Duration:** 6-8 weeks prep - **Level:** Intermediate - **1stepGrow score:** 4/5 - **Best for:** Enterprise engineers in Microsoft-stack organisations **Pros** - Free, high-quality official learning path - Recognised in enterprise procurement and internal banding - Cheap relative to everything else on this list **Cons** - Tests service knowledge, not ML understanding - Expires and needs renewal **Verdict:** Excellent ROI inside a Microsoft shop. Nearly worthless outside one. ### 10. Bootcamp AI/ML Career Tracks Structure and support, sold with claims that need reading carefully. - **Provider:** Various bootcamps - **Price:** ₹1,00,000-₹3,00,000 - **Duration:** 6-9 months - **Level:** Beginner - **1stepGrow score:** 3.6/5 - **Best for:** Career switchers who need accountability and cannot self-direct **Pros** - Live teaching, deadlines and peer cohort genuinely help completion rates - Interview preparation and resume support included **Cons** - Placement guarantees carry eligibility conditions most learners fail to meet - Curriculum depth rarely matches the marketing - Quality varies more than in any other category on this list **Verdict:** Can work for the right person. Demand the placement report with denominators before you pay. ### 11. Kaggle Learn + Competition Track Not a course. Possibly the best portfolio builder there is. - **Provider:** Kaggle - **Price:** Free - **Duration:** Ongoing - **Level:** All levels - **1stepGrow score:** 4.3/5 - **Best for:** Building demonstrable evidence you can actually do the work **Pros** - Micro-courses are short, practical and free - A competition ranking is verifiable evidence in a way a certificate is not - Public notebooks double as a portfolio **Cons** - No curriculum, no structure, no support - Competition skills overlap only partially with production ML **Verdict:** Do this alongside whatever else you pick. A bronze medal beats three certificates in an interview. ## The best AI courses at a glance This table repeats the price, duration and level of each entry above so you can compare them quickly. Prices change often, so confirm the current figure on the provider's page before you commit. | Rank | Course | Price | Duration | Level | |---|---|---|---|---| | 1 | Learnbay GenAI & Agentic AI Master Program | On request (EMI available) | 9 months | Intermediate | | 2 | DeepLearning.AI Deep Learning Specialisation | Subscription (first module free to preview) | 3–4 months | Intermediate | | 3 | Full Stack LLM / Applied AI Engineering | $1,200–$2,500 | 8–10 weeks | Intermediate to advanced | | 4 | fast.ai Practical Deep Learning for Coders | Free | 7–9 weeks | Intermediate | | 5 | Google Cloud Generative AI Learning Path | Free tier, or $29/month | 30–40 hours | Beginner to intermediate | | 6 | IIT / IIM Executive PG in AI & ML | ₹2,50,000–₹4,50,000 | 11–12 months | Beginner to intermediate | | 7 | Hugging Face courses | Free | 20–30 hours | Intermediate | | 8 | MIT / Stanford professional certificates | $1,500–$3,500 | 4–6 months | Advanced | | 9 | Microsoft Azure AI Apps and Agents Developer (AI-103) | $165 exam | 6–8 weeks prep | Intermediate | | 10 | Bootcamp AI/ML career tracks | ₹1,00,000–₹3,00,000 | 6–9 months | Beginner | | 11 | Kaggle Learn and competitions | Free | Ongoing | All levels | ## Which of the best AI courses fits your goal? Start from the job you want, not the course with the loudest marketing. Each goal below points to the entries that serve it best and the trap to avoid. | Your goal | Start with | Add next | Avoid | |---|---|---|---| | Understand how neural networks work | DeepLearning.AI (rank 2) | A modern LLM course | Jumping straight to frameworks | | Ship AI features as a software engineer | Applied AI engineering (rank 3) | Hugging Face courses (rank 7) | Long theory-first programmes | | Learn by doing, for free | fast.ai (rank 4) | Kaggle competitions (rank 11) | Paying for a certificate you do not need | | Work in a Microsoft or Google Cloud shop | AI-103 (rank 9) or Google Cloud (rank 5) | One deployed project on that cloud | Vendor badges for a cloud you do not use | | Research or PhD-adjacent work | MIT / Stanford certificates (rank 8) | Papers and reimplementations | Bootcamps built around tools | | Career switch with no coding yet | Python and statistics first | A structured cohort once you finish something alone | Any course that promises a job in ninety days | ## What do the free AI courses actually give you? The free options give you current, high-quality teaching but no structure, support or credential. Several entries in this ranking cost nothing to start, and the details matter when you choose between them. - **fast.ai** describes [Practical Deep Learning for Coders](https://course.fast.ai/) as a free course for people with some coding experience. Its only stated prerequisites are about a year of programming, preferably in Python, and high-school maths. Part 1 runs to nine lessons of roughly ninety minutes each. - **Hugging Face** offers a free [LLM Course](https://huggingface.co/learn/llm-course/chapter1/1) that began as its NLP course and now centres on large language models. It asks for solid Python and recommends an introductory deep learning course first. - **Google Skills**, the renamed Cloud Skills Boost, has a [free Starter plan](https://www.skills.google/payments/new) with 35 credits a month and a Pro plan at $29 a month for unlimited labs, as of September 2026. - **Kaggle Learn** offers short, practical micro-courses, and its competitions give you a public record that an interviewer can check. The [Deep Learning Specialisation](https://www.coursera.org/specializations/deep-learning) sits just outside this group. It has five courses taught in Python and TensorFlow, and its final course on sequence models includes attention and transformer networks. Since August 2025, however, Coursera has [replaced its free audit option](https://blog.coursera.org/introducing-courseras-new-course-preview-experience/) with a free preview of the first module of most courses, so full access now needs a paid plan or approved financial aid. ## Which AI certification changed in 2026? Microsoft's AI engineer exam changed, so older study guides now point at a retired credential. If a certification is part of your plan, check that it still exists. Microsoft [retired the AI-102 exam](https://learn.microsoft.com/en-us/credentials/support/retired-certification-exams) and the Azure AI Engineer Associate certification on 30 June 2026. The replacement, the [Azure AI Apps and Agents Developer Associate](https://learn.microsoft.com/en-us/credentials/certifications/azure-ai-apps-and-agents-developer-associate/), uses exam AI-103 and covers generative AI and agentic solutions built with Python and Microsoft Foundry. For the wider credential picture, see our ranking of [generative AI certifications](https://www.1stepgrow.com/articles/best-genai-certifications-2026). ## How were these AI courses evaluated? Every programme was scored against five criteria, weighted equally. We set these before looking at any course, and we did not adjust them afterwards. **Curriculum currency.** Does the syllabus reflect how the field works now? A deep learning course that never mentions attention mechanisms is teaching 2018. **Depth versus breadth.** Does it build understanding, or does it tour a list of topics? We looked at assignments, not module titles. A course where you implement backpropagation is doing something a course with a multiple-choice quiz is not. **Honest outcomes.** We checked placement claims against public alumni data. Where a programme claimed a percentage, we asked for the denominator. Several refused to provide it, and the score reflects that. **Cost against the alternative.** Every paid course competes with a free one. The question is not "is this good?" but "is this ₹2,00,000 better than fast.ai plus a Kaggle competition?" **Who it fails.** The most useful thing a review can tell you is who should not buy. Every entry above names that group. ## Common mistakes when choosing an AI course Most regret comes from buying before you can program or before you know what you will build. The learners we interviewed who regretted a purchase mostly made one of these mistakes: - **Starting before they could program.** Every serious course on this list assumes working Python, so the first weeks become a fight with syntax. - **Collecting certificates instead of building.** A certificate gets you a screening call; a project gets you through the technical round. - **Trusting a curriculum page over the assignments.** Ask to see a real assignment, because module titles all look alike. - **Ignoring the renewal and retirement cycle.** Vendor certifications expire or get replaced, as AI-102 did in 2026. - **Buying a long programme that leaves no time to build.** Nine months of evening classes can crowd out the project that actually gets you hired. ## What separated learners who changed careers? Across all 61 learner interviews, one thing separated the people who changed careers from the people who collected certificates: **they built something nobody assigned them.** Not a capstone project from the course template, but something they chose, that solved a problem they cared about, that broke in interesting ways and forced them to read documentation at midnight. Every single person who landed an AI role within twelve months had at least one of these, and most had two. That has an uncomfortable implication for how you should spend your money. A ₹3,00,000 programme that fills every evening for nine months may leave you with no time and no energy to build the thing that actually gets you hired. ## What to do next If you are a working software engineer: learn the fundamentals from DeepLearning.AI, then take an applied LLM engineering course, then ship one real project. The applied course is where most of the cost sits, so compare it against the free Hugging Face material first. If you are a career switcher without a coding background: spend the first three months on Python and statistics before you touch anything on this list. Then start with fast.ai at rank 4, which is free, and see whether you finish it. If you do, you do not need a bootcamp. If you do not, that is genuinely useful information about whether you need one. If your employer is paying: take the expensive one. Get the brand credential on the company's money and spend your own evenings on the project. ## Related comparisons If you are still deciding between fields rather than courses, our [Data Science vs Machine Learning vs AI](https://www.1stepgrow.com/articles/data-science-vs-machine-learning-vs-ai) comparison covers the day-to-day differences, salary bands and which one suits which background. For structured learning paths, see the [AI Engineer roadmap](https://www.1stepgrow.com/articles/ai-engineer-roadmap-2026), and browse every ranked guide on the [lists hub](https://www.1stepgrow.com/lists). ## Frequently asked questions ### Which AI course is best for a complete beginner with no coding background? None of the top three. Spend 8-10 weeks on Python fundamentals and basic statistics first — otherwise you will spend the entire course fighting syntax instead of learning modelling. Once you can write a function and manipulate a DataFrame without help, start with DeepLearning.AI. ### Are expensive AI programmes worth it compared to free ones? You are buying three things: structure, a credential and support. If you finish courses on your own, the free options in this list teach the same material or better. If you have started and abandoned two courses already, the accountability of a paid cohort may genuinely be what you need. ### Do AI certificates actually get you hired? They get you past keyword filters. They do not survive a technical interview. Every hiring manager we spoke to for this piece ranked a working project with a written explanation of the trade-offs above any certificate on the list. Use certificates to get the screening call, then use a project to win the interview. ### How long does it realistically take to become employable in AI? For a working software engineer, six to nine months of consistent evening study. For someone starting without programming experience, twelve to eighteen months. Anyone promising ninety days is selling something. The range depends mostly on whether you already write Python comfortably and how many hours a week you can protect. ### Should I learn machine learning fundamentals or jump straight to LLM APIs? If you want to build AI products, start with the APIs — you will ship something useful in weeks. If you want to be the person a team calls when the model behaves strangely, you need the fundamentals. The second role pays more and is harder to automate. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/best-ai-courses-2026/). Cite with the title, "1stepGrow" and a link._ --- # Which Data Science Course in Singapore Fits You? 8 Options for 2026 > Choosing a data science course in Singapore for 2026? NUS-ISS, NTU, SMU, AIAP and SkillsFuture-eligible bootcamps compared on fees, funding and fit. - **Author:** Ravi Pradhan — AI Expert - 11+ Yrs Experience (https://www.1stepgrow.com/authors/ravi-pradhan/) - **Published:** Aug 28, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 11 min - **Canonical URL:** https://www.1stepgrow.com/articles/best-data-science-ai-courses-singapore-2026/ ## Key takeaways - NUS-ISS stackable graduate certificates are the most flexible university route for working professionals, building towards a Master of Technology in AI Systems. - AI Singapore's AIAP is paid, selective and open only to Singapore citizens, and is one of the most direct routes into AI engineering for people who already code. - Full master's degrees at NTU, NUS and SMU cost roughly S$50,000 to S$70,000 before subsidies, so they make most sense when you need the credential. - For citizens and PRs, SSG subsidies and SkillsFuture Credit can reduce short-course fees substantially, so always compare net cost. The right data science course in Singapore in 2026 depends on your residency and job, not the brand. Working professionals should start with the NUS-ISS stackable AI certificates, citizens who already code should apply to AI Singapore's paid AI Apprenticeship Programme (AIAP), and master's candidates should compare NTU, NUS and SMU, at roughly S$50,000 to S$70,000 before subsidies. Funding is what sets Singapore apart. For citizens and permanent residents, SkillsFuture Singapore subsidies and SkillsFuture Credit can change which course is best value. International professionals usually pay full fees, and for them the credential matters more because of Employment Pass assessments. This guide is for both groups. It ranks eight options, from the NUS-ISS certificates that stack towards a Master of Technology in AI Systems to shorter routes at Heicoders Academy, Vertical Institute and General Assembly. Each entry notes who can use it, and the checklist further down shows how to confirm your net cost before you pay. ## How we compared these courses This is a comparison of published curricula, fees, formats and credentials, based on each provider's official pages and the MySkillsFuture portal as of September 2026. We did not survey learners, count enrolments or interview alumni for this piece, and the ratings are our editorial scores rather than averages of user reviews. We scored each course on six criteria: - **Curriculum depth.** How thoroughly the course covers statistics, machine learning, model evaluation and deployment, and whether generative AI is built on solid fundamentals. - **Format.** Full-time, part-time, online or on campus, and how realistic that is for someone working in Singapore's long-hours finance and tech sectors. - **Credential recognition.** How well the degree or certificate is understood by Singapore employers, and how useful it may be in an Employment Pass application. - **Total cost including subsidies.** The listed fee, the likely net fee after SSG subsidies and SkillsFuture Credit for eligible learners, and the full fee for those who are not eligible. - **Support.** Mentoring, project work, career coaching and industry exposure. - **Fit for working professionals.** Whether someone in a full-time job can realistically complete the course. Where a fee is not published or depends on residency and subsidy eligibility, we give an approximate range or say so. Confirm the current fee on the provider's own page before enrolling. ### 1. Stackable Certificate Programme in Artificial Intelligence (towards MTech in AI Systems) Take one graduate certificate at a time, then stack them into a full NUS master's degree. - **Provider:** NUS-ISS (National University of Singapore) - **Price:** Varies by certificate; SSG subsidies for eligible Singaporeans and PRs (approx.) - **Duration:** Modular; up to 5 years to complete certificates, plus a capstone - **Level:** Intermediate - **1stepGrow score:** 4.5/5 - **Best for:** Working professionals who want an NUS credential without stopping work - **Website:** https://www.iss.nus.edu.sg/stackable-certificate-programmes/stackable-certificate/artificial-intelligence **Pros** - Each graduate certificate is useful on its own and aligned to a specific AI job role - Stacks into a Master of Technology in Artificial Intelligence Systems with a capstone - Part-time format designed for professionals, managers and executives **Cons** - Fees vary by certificate and residency, so the full-degree cost takes some working out - Assumes technical background; not a from-scratch course **Verdict:** The best all-round choice for working professionals in Singapore who want a university credential and can pace their own study over time. ### 2. AI Apprenticeship Programme (AIAP) Paid deep-skilling in AI engineering followed by real industry projects. - **Provider:** AI Singapore - **Price:** No tuition listed; apprentices receive a monthly stipend (S$4,000 listed at time of writing) - **Duration:** 6 or 9 months - **Level:** Intermediate - **1stepGrow score:** 4.5/5 - **Best for:** Singapore citizens who already code in Python and want to become AI engineers - **Website:** https://aiap.sg/apprenticeship/ **Pros** - You are paid while you train, which removes the cost barrier entirely - Three months of deep-skilling followed by work on real industry AI projects - Rigorous selection makes completion a meaningful signal to employers **Cons** - Open only to Singapore citizens with a NITEC, diploma or degree - Competitive, with a take-home technical assessment, technical interview and group case study **Verdict:** If you are eligible and can pass the assessment, this is the strongest value in Singapore. It is a career pivot, not a part-time course. ### 3. Master of Science in Artificial Intelligence (MSAI) A coursework master's in AI from NTU's College of Computing and Data Science. - **Provider:** Nanyang Technological University (NTU) - **Price:** Approx. S$60,000-S$65,000 before subsidies - **Duration:** 1 year full-time; longer part-time - **Level:** Advanced - **1stepGrow score:** 4.4/5 - **Best for:** Graduates with a computing or quantitative background who want a full AI master's - **Website:** https://www.ntu.edu.sg/education/graduate-programme/master-of-science-in-artificial-intelligence **Pros** - Deep technical coverage of machine learning and AI - Full-time and part-time study options - Singaporeans and PRs receive a subsidy on this self-financed programme **Cons** - High fees, especially for international students - Expects strong prior maths and programming **Verdict:** A strong choice when you need a recognised master's degree, including for Employment Pass applications. Overkill if you only need applied skills. ### 4. Master of Computing in Artificial Intelligence A technical AI master's from NUS with coursework and dissertation options. - **Provider:** National University of Singapore (NUS School of Computing) - **Price:** Approx. S$60,000-S$70,000 incl. GST (confirm current fee) - **Duration:** 1.5 years full-time; 2.5 years part-time - **Level:** Advanced - **1stepGrow score:** 4.4/5 - **Best for:** Computing graduates aiming for ML engineering or AI research roles - **Website:** https://www.comp.nus.edu.sg/programmes/pg/mcomp-ai/ **Pros** - Choice between a coursework route and a research dissertation route - Part-time option for people already working in Singapore - NUS is widely recognised across the region **Cons** - One of the most expensive options on this list - Longer than the NTU and SMU programmes **Verdict:** The pick for computing graduates who want research depth or a dissertation. Compare it closely with NTU's MSAI on modules and timing. ### 5. Master of IT in Business (MITB) A business-oriented tech master's with Artificial Intelligence and Data Science & Analytics tracks. - **Provider:** Singapore Management University (SMU) - **Price:** Approx. S$52,000 incl. GST - **Duration:** 12-18 months - **Level:** Intermediate - **1stepGrow score:** 4.3/5 - **Best for:** Professionals who want to apply AI and analytics in business, especially finance - **Website:** https://masters.smu.edu.sg/programme/master-of-it-in-business **Pros** - Specialisation tracks in Artificial Intelligence, Data Science & Analytics, and Financial Technology & Analytics - Strong fit for Singapore's finance and business sectors - Lower total fee than the NTU and NUS master's degrees **Cons** - Less research depth than the NUS and NTU computing degrees - Still a significant cost for self-funded learners **Verdict:** The best master's here for people who want to lead data and AI work inside a business rather than build models full-time. ### 6. AI100 and AI200 (Python Programming and Applied Machine Learning) A Singapore-based, part-time path from Python basics to building machine learning models. - **Provider:** Heicoders Academy - **Price:** From approx. S$900 net after subsidies and credits for eligible learners; full fee higher - **Duration:** Part-time; approx. 4 months for the AI100-AI200 sequence - **Level:** Beginner - **1stepGrow score:** 4.1/5 - **Best for:** Working professionals testing machine learning before a bigger commitment - **Website:** https://heicodersacademy.com/ai200-applied-machine-learning-course **Pros** - Evening and weekend format suits full-time workers - Accepts SkillsFuture Credit, with other funding such as UTAP for eligible learners - Follow-on courses cover deployment and generative AI applications **Cons** - Short courses cannot match university depth - Funding rules differ by course, so check eligibility on MySkillsFuture **Verdict:** A practical, low-risk way to learn applied ML. Good preparation for AIAP or a master's degree. ### 7. Data Science and Data Analytics Courses Short, SkillsFuture-eligible courses that introduce data analytics and data science. - **Provider:** Vertical Institute - **Price:** Varies; up to 70% SSG funding for eligible learners, then SkillsFuture Credit - **Duration:** From approx. 21 hours to multi-week programmes - **Level:** Beginner - **1stepGrow score:** 3.9/5 - **Best for:** Beginners who want a subsidised, low-commitment introduction - **Website:** https://verticalinstitute.com/data-science/ **Pros** - Approved Training Organisation with SSG and IBF funding for eligible learners - Short online format fits around work - Net cost can be low after subsidies and credits **Cons** - Introductory depth only - Not enough on its own to switch into a data scientist role **Verdict:** A sensible first step for a Singaporean or PR testing the waters. Plan to follow it with something deeper. ### 8. Data Analytics and Data Science Bootcamps Intensive bootcamps from an international provider with a Singapore campus. - **Provider:** General Assembly Singapore - **Price:** Varies by format; SkillsFuture subsidies apply to eligible runs - **Duration:** Approx. 12 weeks full-time or 32 weeks part-time (data analytics) - **Level:** Beginner - **1stepGrow score:** 3.8/5 - **Best for:** Career switchers who need an intensive, structured bootcamp with a cohort - **Website:** https://generalassemb.ly/education/data-analytics-bootcamp/singapore **Pros** - Full-time and part-time formats, online or on campus - Some runs are listed on MySkillsFuture with subsidies for eligible learners - Cohort structure and career coaching help people who struggle to self-study **Cons** - Bootcamp depth in statistics and ML is limited compared with a degree - Full-time format means pausing work **Verdict:** Right for a career switcher who needs intensity and structure. Check the specific run's subsidy status before enrolling. ## How do the eight Singapore courses compare? The university master's degrees are the costliest and slowest, AIAP charges no tuition but is open only to citizens, and the short courses and bootcamps are where subsidies make the biggest difference to net cost. | Course | Format | Duration | Indicative fee | Best for | |---|---|---|---|---| | NUS-ISS - Stackable AI certificates (MTech AI Systems) | Part-time, modular | Up to 5 years plus capstone | Varies by certificate; subsidies for eligible learners | Working professionals wanting an NUS degree pathway | | AI Singapore - AIAP | Full-time apprenticeship | 6 or 9 months | No tuition; monthly stipend | Singapore citizens who already code | | NTU - MSc in Artificial Intelligence | Full-time or part-time | 1 year full-time | Approx. S$60,000-S$65,000 before subsidies | Quantitative graduates wanting a full AI master's | | NUS - Master of Computing in AI | Full-time or part-time | 1.5-2.5 years | Approx. S$60,000-S$70,000 | Computing graduates aiming for research or ML roles | | SMU - Master of IT in Business | Full-time or part-time | 12-18 months | Approx. S$52,000 | AI and analytics in business and finance | | Heicoders Academy - AI100/AI200 | Part-time, evenings and weekends | Approx. 4 months | From approx. S$900 net for eligible learners | Testing applied ML | | Vertical Institute - data courses | Short online courses | From approx. 21 hours | Varies; up to 70% SSG funding | Subsidised introductions | | General Assembly Singapore - bootcamps | Full-time or part-time | Approx. 12-32 weeks | Varies by format | Career switchers needing intensity | As of September 2026, all prices are indicative and depend on residency and subsidy eligibility, so confirm the current fee and your net cost on the provider's official page or MySkillsFuture before you pay. ## How do you pick a data science course in Singapore? **Work out your net cost first.** If you are a Singapore citizen or PR, SSG course fee subsidies on approved courses can cover a large share of the fee, with higher support for eligible mid-career citizens. Citizens can then use [SkillsFuture Credit](https://www.myskillsfuture.gov.sg/content/portal/en/career-resources/career-resources/education-career-personal-development/SkillsFuture_Credit.html), including the SkillsFuture Credit (Mid-Career) for those aged 40 and above, on eligible courses. Union members may be able to use NTUC's UTAP funding as well. Check your balance and each course's funding status on MySkillsFuture, because two courses with the same listed fee can have very different net costs. **Consider the SkillsFuture Career Transition Programme if you are switching fields.** SCTP courses are longer and more job-oriented than typical short courses, and are aimed at mid-career Singaporeans and PRs moving into a new sector. Several data and AI courses are offered under the scheme. **Decide whether you need a credential or a portfolio.** For many local analyst and engineering roles, demonstrable skills and projects matter more than the certificate. A recognised master's degree matters more for research roles, for some finance and public-sector positions, and for international professionals, since Employment Pass assessments under the COMPASS framework consider qualifications. Check [MOM's current criteria](https://www.mom.gov.sg/passes-and-permits/employment-pass/eligibility) rather than relying on second-hand summaries. **Pick projects that match local employers.** Singapore's biggest data employers are in banking and finance, regional tech headquarters, government digital services, logistics and healthcare. A fraud detection, credit risk or demand forecasting capstone speaks directly to them. Because regulators and public bodies emphasise responsible AI, showing that you can explain and evaluate a model is a real advantage. **Be realistic about time.** Singapore's working hours are long in finance and tech. Part-time evening and weekend formats, or modular routes such as NUS-ISS, are usually more sustainable than squeezing a full-time course around a job. Full-time options such as AIAP and the university master's degrees work best when you can step away from work. **Check the specific course run.** Funding in Singapore is attached to particular courses and often to particular runs, identified by a course reference number on MySkillsFuture. A provider may offer the same subject in a funded and an unfunded version, or change which runs qualify. Before you pay, confirm that the exact run you are joining is eligible for the subsidy and credit you expect, and keep a record of it. **Ask your employer.** Many Singapore employers sponsor training, particularly in finance and the public sector, and some courses qualify for employer-side funding as well. A part-time master's or graduate certificate that your employer co-funds can be far better value than a self-funded bootcamp, even if it takes longer. **Questions to ask before you pay.** Whichever provider you are considering, get clear written answers to these: - What is my net fee after SSG subsidies, and how much SkillsFuture Credit can I apply to it? - Is this specific course run eligible for the funding shown, and what attendance or assessment conditions apply? - Who teaches the sessions, and what is their industry experience? - What does career support include, and is it available to international learners as well as citizens and PRs? - Which projects will I be able to show employers, and do they use real or realistic data? - If I miss sessions because of work, can I catch up or join a later run? In our experience of hiring for data and AI teams, candidates who can walk through one well-built project in depth do better than those holding several short certificates, so weigh the project work heavily in your decision. ## Which course should you choose? **You are a working professional in tech or finance and want a university credential.** Start with an NUS-ISS graduate certificate in AI. You can stop after one certificate if it gives you what you need, or keep stacking towards the Master of Technology in AI Systems. **You are a Singapore citizen who already codes and wants to become an AI engineer.** Apply to AIAP. It is competitive, but you are paid while you train and you finish with real project experience. If you are not yet ready for the assessment, take Heicoders Academy's AI100 and AI200 first. **You have a computing or quantitative degree and want a full master's.** Compare NTU's MSc in Artificial Intelligence and NUS's Master of Computing in AI on modules, part-time options and timing. Choose NUS if you want a dissertation option; NTU if you want a shorter full-time year. **You want to lead data and AI work inside a business, especially in finance.** SMU's Master of IT in Business, with its Artificial Intelligence or Data Science & Analytics tracks, is built for this and costs less than the computing master's degrees. **You are a Singaporean or PR testing whether data science suits you.** Take a short, subsidised course from Vertical Institute or Heicoders Academy using SkillsFuture Credit. The net cost can be low, and you will know within weeks whether to go further. **You are switching careers and need intensive structure.** A General Assembly bootcamp or an SCTP-funded course gives you a cohort, deadlines and career coaching. Check the run's subsidy status and plan to deepen your statistics afterwards. **You are an international professional moving to Singapore.** Budget for full fees, since subsidies usually do not apply. A recognised master's from NUS, NTU or SMU can support both your job search and an Employment Pass application. ## Related reading For the local job market, funding schemes and work passes, see our [Singapore region hub](https://www.1stepgrow.com/regions/singapore/). To compare options globally, read [the best AI courses in 2026](https://www.1stepgrow.com/articles/best-ai-courses-2026). If you are deciding on a role, our [AI engineer roadmap for 2026](https://www.1stepgrow.com/articles/ai-engineer-roadmap-2026) sets out the skills, and [Is a data science bootcamp worth it?](https://www.1stepgrow.com/articles/is-a-data-science-bootcamp-worth-it) covers the bootcamp question in more depth. ## Frequently asked questions ### What is the best data science course in Singapore? For working professionals, NUS-ISS stackable graduate certificates offer the best balance of credential and flexibility. Singapore citizens who already code should apply to AI Singapore's AIAP, and those who need a full master's should compare NTU's MSAI, NUS's MComp in AI and SMU's MITB. ### How much does a data science course in Singapore cost? Short SkillsFuture-eligible courses can cost a few hundred to a few thousand dollars after subsidies for eligible learners. Full master's degrees at NTU, NUS and SMU cost roughly S$50,000 to S$70,000 before subsidies. AIAP charges no tuition and pays a stipend. ### Can I use SkillsFuture Credit for data science courses? Yes, for courses listed as SkillsFuture Credit-eligible on the MySkillsFuture portal. Citizens aged 25 and above receive an opening credit, and those aged 40 and above can also receive the SkillsFuture Credit (Mid-Career). SSG course fee subsidies are usually applied first, and credit covers some or all of the rest. ### What is the SkillsFuture Career Transition Programme? It is a SkillsFuture Singapore scheme that funds longer, industry-oriented courses for mid-career Singaporeans and PRs moving into a new field. Several data and AI courses are offered under it by approved providers, often with career support. ### Can international students or foreigners take these courses? Yes, for university programmes and most private courses, but at full fees because SkillsFuture subsidies target citizens and PRs. AIAP is open only to Singapore citizens. Working in Singapore afterwards usually requires an Employment Pass, so check MOM's current criteria. ### Is a master's degree necessary for a data science job in Singapore? No. Many employers hire on skills and experience, especially for analyst and engineering roles. A recognised master's matters more for research roles and for international applicants whose Employment Pass assessment considers qualifications. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/best-data-science-ai-courses-singapore-2026/). Cite with the title, "1stepGrow" and a link._ --- # Advanced Data Science and AI Course: What Really Separates Them > Choosing an advanced data science and AI course? Five options compared on sequencing, project depth and price, plus four questions to ask before you pay. - **Author:** Nishant Kiran — B2B SaaS & EdTech Content Strategist (https://www.1stepgrow.com/authors/nishant-kiran/) - **Published:** Aug 27, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/advanced-data-science-ai-course/ ## Key takeaways - The differentiator in 2026 is no longer whether a course covers deep learning — it is whether generative AI is integrated into the curriculum or bolted onto the end. - Programs built for working professionals need live cohorts and flexible batches, because self-paced study leaves all of the accountability to you. - Project depth beats module count. Four end-to-end capstones with real data beat forty recorded hours. - For non-programmers, an advanced program is the wrong starting point regardless of price. - As of September 2026, the US Bureau of Labor Statistics puts median data scientist pay at $120,230 (May 2025); no official Indian source publishes data science salary bands. An advanced data science and AI course is worth paying for when it teaches statistics and machine learning before generative AI, makes you take messy data to a defended conclusion several times, and fits around a working week. Course marketing has converged on the same promises (expert mentors, placement assistance, capstones), so it carries no information. This comparison looks past that copy at what actually varies: curriculum sequence, project depth, who teaches, and which learner each programme suits. It is for people who can already program and want depth rather than a survey. It ranks five options, from a 13-month live cohort to free fast.ai lessons, and ends with four questions to ask any provider before you enrol. ### 1. Learnbay — Advanced Data Science & GenAI Master Program Statistics and ML fundamentals first, generative AI layered on top — structured around a full-time job. - **Provider:** Learnbay - **Price:** On request (EMI available) - **Duration:** 13 months - **Level:** Intermediate - **1stepGrow score:** 4.6/5 - **Best for:** Working professionals targeting senior data science roles - **Website:** https://www.learnbay.co/datascience/data-science-and-ai-master-certification-program **Pros** - Fundamentals sequenced before the GenAI layer, so the AI modules build on something - Four capstone projects with IBM project certification rather than notebook exercises - Domain electives — BFSI, healthcare, retail, manufacturing — so projects match your industry - Live cohort delivery with weekend batches designed around a working week **Cons** - Thirteen months is a long commitment; shorter options exist if you only need the GenAI layer - Pricing is not published — you have to request it **Verdict:** The most complete option here for a working professional who wants depth rather than a survey course. The sequencing is the real advantage: too many programs teach LLM tooling to people who never learned to validate a model. ### 2. DeepLearning.AI — Deep Learning Specialisation Still the clearest explanation of neural network fundamentals available at any price. - **Provider:** Coursera - **Price:** $49/month (first module free to preview) - **Duration:** 3-4 months - **Level:** Intermediate - **1stepGrow score:** 4.8/5 - **Best for:** Self-directed learners who want fundamentals cheaply - **Website:** https://www.coursera.org/specializations/deep-learning **Pros** - First module free to preview; financial aid available for full access - Assignments build architectures from scratch before touching a framework - Universally recognised by hiring managers **Cons** - Transformers appear only in the final course (added in a 2021 update), so it still needs pairing with a dedicated LLM course - No mentorship, no cohort, no placement support **Verdict:** Unbeatable value for fundamentals. Not a career program — there is no support structure and no credential weight beyond the name. ### 3. IIT-partnered Executive PG Programs The brand-name option, priced accordingly. - **Provider:** Various institutes via EdTech partners - **Price:** ₹2,50,000-₹4,50,000 - **Duration:** 11-12 months - **Level:** Beginner to intermediate - **1stepGrow score:** 3.9/5 - **Best for:** Professionals who need an institutional credential for internal promotion **Pros** - Institute branding clears HR filters, especially in Indian corporate hiring - Structured cohort and deadlines improve completion rates - Alumni networks carry genuine weight in some sectors **Cons** - Teaching is frequently delivered by the EdTech partner, not institute faculty - Curriculum often lags the field by a year or more - Placement support quality varies dramatically between cohorts **Verdict:** Buy it for the credential and the accountability, not the curriculum. Verify who actually teaches before signing. ### 4. Fast.ai — Practical Deep Learning for Coders Top-down teaching that gets you to a working model in lesson one. - **Provider:** fast.ai - **Price:** Free - **Duration:** 7-9 weeks - **Level:** Intermediate - **1stepGrow score:** 4.7/5 - **Best for:** Coders who hate theory-first courses - **Website:** https://course.fast.ai/ **Pros** - Completely free with no certificate paywall - You train a working model in the first session - Unusually good community forum **Cons** - No credential - Library abstractions hide things interviewers will ask about **Verdict:** The best free deep learning course there is. Do it alongside a structured program, not instead of one, if you need a credential. ### 5. University Professional Certificates Academic rigour, academic pace, academic price. - **Provider:** edX / Stanford Online - **Price:** $1,500-$3,500 - **Duration:** 4-6 months - **Level:** Advanced - **1stepGrow score:** 4.1/5 - **Best for:** Researchers and engineers heading toward specialised work **Pros** - Genuine mathematical depth - Globally recognised without qualification **Cons** - Slow to cover applied and generative topics - Poor value for applied engineering roles **Verdict:** Right for research-track careers, expensive overkill for product work. ## How do these advanced programmes compare side by side? The summary below uses the figures from the list above, as of September 2026. Fees and batch formats change, so treat the provider's own page as the final word. Coursera also [replaced free auditing with a free first-module preview](https://blog.coursera.org/introducing-courseras-new-course-preview-experience/) in August 2025, so finishing a Coursera specialisation now needs a paid plan or approved financial aid. | Programme | Delivery | Price | Duration | Credential weight | Best for | |---|---|---|---|---|---| | Learnbay Advanced Data Science & GenAI Master Program | Live cohort, weekend batches | On request | 13 months | Programme and IBM project certificates | Working professionals wanting depth | | DeepLearning.AI Deep Learning Specialisation | Self-paced | $49/month (first module free to preview) | 3-4 months | Well-known name, no support | Self-directed fundamentals | | IIT-partnered Executive PG Programs | Cohort via EdTech partner | ₹2,50,000-₹4,50,000 | 11-12 months | Institute brand | Credential-driven promotions | | fast.ai Practical Deep Learning for Coders | Free video lessons | Free | 7-9 weeks | None | Coders who learn by building | | University professional certificates | Academic, part-time | $1,500-$3,500 | 4-6 months | Strong academic name | Research-track careers | The pattern is clear: the free and cheap options are excellent on content but offer no accountability. The expensive ones sell structure, a cohort and a credential. Decide which of those you are actually short of before comparing prices. ## What actually differentiates an advanced data science and AI course in 2026? **Where generative AI sits in the sequence.** This is the single biggest curriculum difference right now. Some programs teach statistics, machine learning and validation first, then build the generative layer on top of it. Others append three GenAI modules to a 2022 syllabus. The difference shows up in graduates. Someone who learned evaluation before they learned prompting can tell you whether their retrieval system got better. Someone who did it the other way round ships things nobody can measure. **Project depth versus module count.** A syllabus with forty modules sounds thorough and usually is not. What matters is how many times you take a messy dataset to a defended conclusion. Four genuine end-to-end projects teach more than forty recorded hours, and they are what you actually talk about in interviews. **Whether it was designed for people with jobs.** Self-paced study puts the whole burden of finishing on you, and a demanding job wins most weeks. Live cohorts, fixed batch times and weekend schedules exist because accountability is the binding constraint for this audience — not content availability. ## What does the data scientist role actually ask for? Check the job before you choose the course. The US Bureau of Labor Statistics' [occupational profile for data scientists](https://www.bls.gov/ooh/math/data-scientists.htm) lists duties such as collecting and analysing data, creating, validating and testing models, and making business recommendations to stakeholders. It also puts median US pay at $120,230 in May 2025, with a bachelor's degree as the typical entry-level education. Two cautions apply. First, those are US figures. No Indian official source publishes data science salary bands, so treat any rupee salary claim in a course brochure as marketing unless it names its source. Second, notice how much of the duty list is validation and communication rather than modelling. An advanced course that skips both is teaching half the job. ## Which credential is worth paying for? Credentials carry different weight depending on where you want to work: - **An institute brand** helps most with HR screening and internal promotion committees, especially in Indian corporate hiring. - **A well-known online specialisation** signals that you studied the fundamentals, but it rarely moves a hiring decision alone. - **A public portfolio** — repositories, a written evaluation, a deployed demo — is what technical interviewers actually probe. In practice, the credential opens the door and the portfolio carries the interview. Therefore, pay for a certificate only when a specific employer or promotion process asks for it. ## What should you know before starting an advanced course? You need working Python and basic statistics, including inference. Which of those you already have decides whether an advanced programme is the right level. If you can already program and know basic statistics, an advanced program is the right level and you should pick based on project depth and domain fit. If you can program but statistics is shaky, do not skip it. Every advanced curriculum assumes inference, and the people who struggle hardest are those who can build a model but cannot say whether the result is real. If you cannot program yet, none of these are for you today. Spend eight to ten weeks on Python and basic statistics first — our [foundation course guide](https://www.1stepgrow.com/articles/data-science-course-for-non-programmers) covers that route specifically. Starting at the advanced level without that base is the most common and most expensive mistake in this market. Even the free options expect a base. For example, fast.ai says its [Practical Deep Learning course](https://course.fast.ai/) is for people with about a year of coding experience, preferably in Python. ## Who should skip an advanced data science and AI course? Some readers will get more from a narrower purchase: 1. **You only need the generative AI layer.** If you already model well, a focused certification is cheaper; see our guide to [generative AI certifications worth taking](https://www.1stepgrow.com/articles/best-genai-certifications-2026). 2. **You want research roles.** A university certificate or a master's degree fits that path better than an applied bootcamp. 3. **You have not yet written Python.** Start with the foundation route above. ## What should you ask a provider before enrolling? Ask four things: placement figures with denominators, the names of the instructors, a recent capstone repository, and what happens when you fall behind. Start with the placement report with denominators: how many enrolled, how many were eligible under the terms, how many were placed. Percentages without a denominator are marketing. Ask who teaches. Get names and look them up, particularly for institute-branded programs where delivery is often handled entirely by the EdTech partner. Ask to see a capstone project from the last cohort. If the answer is a slide deck rather than a repository, adjust your expectations about project depth. Ask what happens if you fall behind. Working professionals miss weeks — a program with batch-repeat access handles that, and one without it quietly writes you off. Finally, check how current the deep learning material is. The Deep Learning Specialisation, for instance, notes on its [Coursera page](https://www.coursera.org/specializations/deep-learning) that an April 2021 update added Transformers to its Sequence Models course. Any syllabus should be able to tell you when it was last revised. ## Related reading If you are still choosing between fields rather than programs, read [Data Science vs Machine Learning vs Deep Learning vs AI](https://www.1stepgrow.com/articles/ai-vs-machine-learning-vs-deep-learning). For structured self-study, our [data scientist roadmap](https://www.1stepgrow.com/articles/data-scientist-roadmap-2026) lays out the same material as a stage-gated path, and [Coursera vs Udemy vs edX](https://www.1stepgrow.com/articles/coursera-vs-udemy-vs-edx) compares the platforms several of these courses run on. ## Frequently asked questions ### What makes a data science course 'advanced' rather than introductory? Three things: it assumes you can already program and handle basic statistics, it requires you to make modelling decisions rather than follow instructions, and its projects use messy real data rather than pre-cleaned datasets. If the syllabus starts with 'introduction to Python', it is not an advanced course regardless of the title. ### Should generative AI be part of an advanced data science course in 2026? Yes, but sequenced correctly. Courses that teach LLM tooling to people who cannot validate a model produce graduates who ship systems nobody can evaluate. Look for programs where statistics and ML come first and the generative layer builds on them. ### How long should an advanced data science program take? For a working professional studying part-time, nine to thirteen months is realistic for genuine depth. Programs claiming to make you job-ready in three months are either assuming substantial prior experience or overstating the outcome. ### Do I need an advanced course if I already work with data? If you are already doing analysis professionally, the gap is usually machine learning depth and deployment rather than fundamentals. A focused program is better value than a broad one — and your employer may fund it, which changes the economics entirely. ### Are these programs suitable for non-programmers? No. Every option on this list assumes working programming ability. Non-programmers should start with a foundation course and move up — attempting an advanced program without that base is the single most common reason people drop out. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/advanced-data-science-ai-course/). Cite with the title, "1stepGrow" and a link._ --- # Data Science Course for Non-Programmers: What Really Starts at Zero > Most beginner courses assume you can code by week two. How to pick a data science course for non-programmers, the real timeline, and the first job to target. - **Author:** Nishant Kiran — B2B SaaS & EdTech Content Strategist (https://www.1stepgrow.com/authors/nishant-kiran/) - **Published:** Aug 26, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 8 min - **Canonical URL:** https://www.1stepgrow.com/articles/data-science-course-for-non-programmers/ ## Key takeaways - Most 'beginner friendly' data science courses assume programming experience they never state. Check the first module, not the marketing. - A genuine non-programmer path needs 8-12 weeks of Python and statistics before any modelling content. - Twelve to eighteen months to employability is the honest timeline from a standing start. - Analytics roles are a more achievable first target than data scientist, and a well-trodden route into the field. A genuine data science course for non-programmers spends its first weeks on Python basics and statistics before any modelling, sets a realistic timeline of twelve to eighteen months to employability, and points you at analyst roles first. Most "beginner friendly" courses fail that test, so check the first module before you pay. Almost every "beginner friendly" data science course has a quiet assumption buried in week two: that you can already write a loop. The marketing says no experience required; the first assignment says `import pandas as pd`. Many people who hit that wall conclude they are bad at this, when they were simply sold a course written for someone else. This guide is for career switchers and non-technical professionals starting without code. It compares five options that genuinely assume nothing, gives a five-point test for any syllabus, lays out the month-by-month timeline nobody puts on a landing page, and explains why analyst is usually the better first job title. ### 1. Learnbay — Applied AI Practitioner Program Explicitly built for non-programmers, with IIT Patna certification and a four-month commitment. - **Provider:** Learnbay - **Price:** On request - **Duration:** 4 months - **Level:** Beginner - **1stepGrow score:** 4.4/5 - **Best for:** Non-technical professionals who need applied AI literacy without a career break - **Website:** https://www.learnbay.co/artificial-intelligence/applied-ai-practitioner-program **Pros** - No coding prerequisite — the curriculum is designed around that constraint rather than apologising for it - IIT Patna certification carries weight in Indian corporate hiring - Four months is short enough to finish alongside a demanding job - Two capstone projects rather than quizzes **Cons** - Applied rather than deep — this is literacy and practice, not a research foundation - You will still need further study to move into a full data science role **Verdict:** The most realistic entry point on this list for someone genuinely starting from zero. It solves the actual problem non-programmers have — every other course assumes a base they do not have. ### 2. Learnbay — Data Analytics Certification Program Longer, deeper, and aimed at the job title non-programmers can realistically reach first. - **Provider:** Learnbay - **Price:** On request (EMI available) - **Duration:** 9 months - **Level:** Beginner - **1stepGrow score:** 4.5/5 - **Best for:** Non-tech professionals targeting an actual analytics job - **Website:** https://www.learnbay.co/datascience/data-analytics-certification-course **Pros** - Foundation programming classes included rather than assumed - 275 hours covering live lectures, practice and interview preparation - Weekday and weekend batches; subscription access to repeat batches for 2-3 years - Targets analyst roles, which are far more numerous than data scientist roles **Cons** - Nine months is a real commitment - Analytics is a lower ceiling than data science without further study **Verdict:** If your goal is employment rather than literacy, this is the better of the two. Analyst is the realistic first job title from a non-technical start. ### 3. Python for Everybody The gentlest genuinely good introduction to programming there is. - **Provider:** Coursera (University of Michigan) - **Price:** $49/month (first module free to preview) - **Duration:** 6-8 weeks - **Level:** Absolute beginner - **1stepGrow score:** 4.7/5 - **Best for:** Building the programming base every other course assumes - **Website:** https://www.coursera.org/specializations/python **Pros** - Genuinely assumes nothing - First module free to preview; financial aid available - Excellent pacing for people who have never written code **Cons** - Programming only — no statistics, no modelling - No support structure **Verdict:** Do this first, whatever else you buy. It costs nothing and makes every subsequent course easier. ### 4. IBM Data Science Professional Certificate Broad, beginner-friendly, and honest about being an introduction. - **Provider:** Coursera - **Price:** $49/month - **Duration:** 3-5 months - **Level:** Beginner - **1stepGrow score:** 4.1/5 - **Best for:** A low-risk first commitment - **Website:** https://www.coursera.org/professional-certificates/ibm-data-science **Pros** - Gentle on-ramp with no prerequisites - Covers the full workflow end to end - Cheap enough to abandon without regret **Cons** - Shallow — introduces topics rather than teaching them - Very widely held, so weak as a differentiator **Verdict:** A good first step and a poor last one. ### 5. Khan Academy Statistics & Probability Free, and the gap most career switchers do not know they have. - **Provider:** Khan Academy - **Price:** Free - **Duration:** 8-10 weeks - **Level:** Beginner - **1stepGrow score:** 4.6/5 - **Best for:** Everyone on this path, in parallel with programming - **Website:** https://www.khanacademy.org/math/statistics-probability **Pros** - Free and self-paced - Fixes the most common weakness in bootcamp graduates - Directly improves interview performance **Cons** - No credential or structure **Verdict:** Unglamorous and the highest-return free thing you can do. ## How do the options for non-programmers compare? The table below summarises the list, with prices and durations as of September 2026. Because fees and batch formats change, confirm them on each provider's page. Note that Coursera [replaced free auditing with a free first-module preview](https://blog.coursera.org/introducing-courseras-new-course-preview-experience/) in August 2025, so completing its courses now needs a paid plan or approved financial aid. | Option | Coding assumed? | Price | Duration | What you come out with | |---|---|---|---|---| | Learnbay Applied AI Practitioner Program | None | On request | 4 months | Applied AI literacy and two capstones | | Learnbay Data Analytics Certification Program | None; foundation classes included | On request | 9 months | Analyst-level skills and interview preparation | | Python for Everybody (University of Michigan) | None | $49/month (first module free to preview) | 6-8 weeks | A programming base | | IBM Data Science Professional Certificate | None | $49/month | 3-5 months | A broad introduction to the workflow | | Khan Academy Statistics & Probability | None | Free | 8-10 weeks | The statistics most switchers lack | The cheapest rows are also the best first steps. Coursera's page for [Python for Everybody](https://www.coursera.org/specializations/python) describes it as an on-ramp for programming with no prerequisites. Similarly, the [IBM Data Science Professional Certificate](https://www.coursera.org/professional-certificates/ibm-data-science) states that no prior programming experience is necessary, although it notes that high school maths helps. ## How can you tell whether a course is really for non-programmers? Landing pages all claim to welcome beginners, so test the syllabus instead: 1. **Read the week-one topics.** Variables, loops and functions mean the course starts from zero. By contrast, "Introduction to pandas" means it assumes you can already code. 2. **Count the weeks before modelling.** A genuine non-programmer path spends eight to twelve weeks on Python and statistics first. 3. **Check whether statistics is taught or assumed.** Hypothesis testing and confidence intervals should appear as lessons, not as prerequisites. 4. **Ask what support you get when stuck.** Beginners get stuck on setup and syntax far more than on concepts, so live doubt-clearing matters more at this level. 5. **Look at the capstone.** It should use a dataset you prepare yourself rather than a pre-cleaned file. Even free material needs this check. For instance, Python's own [official tutorial](https://docs.python.org/3/tutorial/index.html) says plainly that it is written for programmers new to Python, not for beginners new to programming. ## What is the realistic timeline from zero? From a genuine standing start, here is what the sequence actually looks like. **Weeks 1-10: programming and statistics, in parallel.** Python fundamentals until you can write a function and manipulate a dataset without following along. Statistics until distributions and hypothesis testing make sense. This stage feels slow and it is not optional — everything downstream assumes it. **Months 3-9: a structured program.** This is where a paid course earns its money, because you now have the base to use it. Accountability, feedback on your code, and a cohort matter more at this stage than content, which is freely available. **Months 9-15: portfolio and applications.** Two projects you scoped yourself, then a tracked application pipeline. Our [learning-to-earning roadmap](https://www.1stepgrow.com/articles/learning-to-earning-tech-career-roadmap) covers this stage specifically. Twelve to eighteen months, part-time, with income throughout if you do not quit your job. That is the honest number. ## Which job title should you aim for first? The advice non-programmers usually get is to target data scientist, because it pays more. That advice ignores the thing that actually determines outcomes: how many doors are open to you. Analyst roles typically screen less severely for a first hire, and they put you inside an organisation with real data. From there, the internal move into data science skips the hardest filter in the whole process — the external hiring bar for a first data science job. The official description of the destination role explains why. The US Bureau of Labor Statistics' [data scientist profile](https://www.bls.gov/ooh/math/data-scientists.htm) lists creating, validating and testing models alongside the analysis work, and gives a bachelor's degree as the typical entry-level education. That is a high bar to clear from a standing start, but a reachable one after a few years as an analyst. Our [analyst vs scientist comparison](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist) has the full breakdown, but the short version: analyst is usually the better door, even when data scientist is the better destination. ## Who should not take a data science course for non-programmers? This route is not right for everyone. Consider something else if: - **You already code in any language.** Start at an intermediate level instead; our guide to [advanced data science and AI courses](https://www.1stepgrow.com/articles/advanced-data-science-ai-course) covers that. - **You only need to manage or buy AI work.** A decision-focused programme fits better than a technical one. - **You cannot commit a few hours every week for a year.** In that case, the free options let you test your interest before spending money. ## How should you present a non-technical background? Lead with your domain expertise and treat the technical skills as the addition. The most common mistake non-programmers make is presenting themselves as beginners. You are not a beginner. Rather, you are someone with years of domain expertise adding technical skill. A logistics coordinator who learns SQL and Python is not a junior analyst — they are an analyst who already understands why the delivery data is wrong in that specific way, which takes a fresh graduate two years to learn. Every industry has this. The strongest candidates from non-technical backgrounds lead with the domain and treat the technical skills as the addition, not the identity. ## Related reading For the level above this one, see our guide to [advanced data science and AI courses](https://www.1stepgrow.com/articles/advanced-data-science-ai-course). If you are weighing a bootcamp, [is a data science bootcamp worth it](https://www.1stepgrow.com/articles/is-a-data-science-bootcamp-worth-it) covers the economics honestly. ## Frequently asked questions ### Can I really learn data science with no programming background? Yes, but not on the timeline most marketing implies. Budget eight to twelve weeks purely on Python and statistics before any modelling content, and twelve to eighteen months total before you are competitive for roles. People who compress this are the ones who drop out. ### How do I tell whether a course actually welcomes non-programmers? Read the first module, not the marketing page. If week one is 'Introduction to pandas' rather than 'variables and loops', the course assumes programming ability it never declared. Genuine non-programmer courses spend real weeks on fundamentals. ### Should I aim for data scientist or data analyst first? Analyst, in almost every case. Analyst roles typically screen less severely for a first hire, they put you inside an organisation with real data, and moving internally from analyst to data scientist avoids the hardest external hiring bar in the field. ### Do I need to be good at maths? You need comfort with statistics — distributions, inference, what a confidence interval means. You do not need calculus or linear algebra at the level most people fear. The maths that matters is the maths that decides whether a result is real. ### Is my non-technical background a disadvantage? It is an advantage if you position it correctly. Domain knowledge in insurance, healthcare, logistics or retail takes a graduate years to acquire. The strongest non-programmer candidates present themselves as domain experts adding technical skill, not as beginners. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/data-science-course-for-non-programmers/). Cite with the title, "1stepGrow" and a link._ --- # AI vs Machine Learning vs Deep Learning (and Where Data Science Fits) > AI vs machine learning vs deep learning, plus data science: how the four terms nest, one problem solved four ways, and when deep learning is worth using. - **Author:** Althaf Ashraf — AI Systems Engineer, Tata Consultancy Services (https://www.1stepgrow.com/authors/althaf-ashraf/) - **Published:** Aug 25, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Comparison · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/ai-vs-machine-learning-vs-deep-learning/ ## Key takeaways - They are nested, not parallel: AI is the field, machine learning is an approach within it, deep learning is a technique within that, and data science is a discipline that overlaps all three. - Job titles use these words interchangeably. Read the responsibilities section, never the title. - Deep learning is the narrowest and most over-taught of the four relative to the number of jobs that need it. - For most people the practical choice is between data science and applied AI engineering — not between all four. AI vs machine learning vs deep learning is not a contest between rivals, because the three nest. AI is the whole field, machine learning is one approach within it, and deep learning is one family of machine learning techniques. Data science is different again: a discipline that overlaps all three but sits inside none of them. Getting that structure right matters, because the wrong mental model leads to the wrong study plan: usually months spent on deep learning by someone who needed statistics and SQL. This guide is for learners and career changers trying to make sense of the terms. It shows how they nest in one table, solves a single spam-filtering problem four ways, and sets out when deep learning is worth its cost. If you have already decided you want a job in this space and need to choose a role, the [data science vs machine learning vs AI career comparison](https://www.1stepgrow.com/articles/data-science-vs-machine-learning-vs-ai) covers the working weeks, skills and pay in more depth. ## AI vs machine learning vs deep learning: how do they nest? Think of three circles, one inside the next, with data science as a fourth circle overlapping them. **Artificial intelligence** is the broad field: making machines do things that look intelligent. It is the oldest and widest term. John McCarthy and colleagues used it in their 1955 [proposal for the Dartmouth summer research project](https://www-formal.stanford.edu/jmc/history/dartmouth/dartmouth.html), which set out to study how machines could use language, form abstractions and improve themselves. AI also includes rule-based systems that involve no learning at all. **Machine learning** is one approach within AI. Rather than writing the rules yourself, you let an algorithm infer patterns from examples. Most of what industry calls AI today is machine learning. **Deep learning** is a family of techniques within machine learning. In the words of the landmark [2015 Nature review by LeCun, Bengio and Hinton](https://pubmed.ncbi.nlm.nih.gov/26017442/), it uses models "composed of multiple processing layers" that learn representations of data. Deep learning drives most recent visible progress — language models, image generation, speech — which is why people use "deep learning" and "AI" interchangeably even though one sits two levels inside the other. **Data science** is the odd one out. It is not a subset of AI; instead, it is a discipline defined by its purpose rather than its technique. A data scientist uses statistics, SQL, visualisation and sometimes machine learning to answer questions. Plenty of excellent data science involves no ML at all. | Term | What it is | Defining idea | Typical example | |---|---|---|---| | Artificial intelligence | The field | Machines performing tasks that look intelligent | A route planner, a chess engine, a chatbot | | Machine learning | An approach within AI | Learning patterns from data instead of hand-written rules | A credit-risk model trained on past loans | | Deep learning | A technique within ML | Many-layered neural networks that learn their own features | Speech recognition, image classification, LLMs | | Data science | A discipline that overlaps all three | Using data to support a decision | An A/B test deciding which checkout page to keep | | Feature | Data Science | Machine Learning | Deep Learning / AI | | --- | --- | --- | --- | | What it actually is | A discipline — questions, statistics, communication | An approach — algorithms that learn from data | A technique (DL) inside the broader field (AI) | | Relationship | Overlaps all three | A subset of AI | DL is a subset of ML; AI is the umbrella | | Core daily work | SQL, analysis, experiments, stakeholder decks | Feature engineering, training, deployment, monitoring | Model integration, retrieval, evaluation, fine-tuning | | Maths depth needed | High — inference and statistics | Medium — applied | Medium for applied, very high for research | | Software engineering depth | Low to medium | High | Medium to high | | Volume of open roles | High | High | Growing fastest | | Entry difficulty from non-tech | Moderate | Hard | Easiest via the applied AI route | | Median India salary (3-5 yrs) | ₹18 LPA | ₹26 LPA | ₹22 LPA | | Automation exposure | Medium — routine reporting | Low | Medium — tooling commoditising fast | | Portfolio that gets interviews | An analysis that changed a decision | A deployed model with monitoring | A working AI product with evaluation | **Verdict — Machine Learning:** Taken purely on pay, demand and resistance to automation, machine learning is the strongest of the three tracks in 2026 — but it is also the hardest to enter, because it demands competence in two disciplines rather than excellence in one. Deep learning is best understood as a technique you learn within ML or AI work, not as a career target of its own. For most readers the real decision is between data science and applied AI engineering. ## How would each approach build a spam filter? Rule-based AI blocks known phrases, machine learning learns which words matter from labelled emails, deep learning learns its own representation of the text, and data science asks whether the filter was worth building. Watching each tackle the same problem makes the difference concrete. 1. **Rule-based AI.** An engineer writes rules: block any email that contains certain phrases or comes from a known bad domain. It is AI in the classic sense, yet nothing is learned, so spammers only need to change their wording. 2. **Machine learning.** You collect thousands of emails already labelled "spam" or "not spam", count the words in each, and train a classifier such as logistic regression. The model learns which words matter, and it adapts when you retrain on new examples. 3. **Deep learning.** Instead of hand-built word counts, a neural network reads the raw text and learns its own representation of meaning. As a result, it can catch spam that avoids every obvious keyword, at the cost of more data, more compute and less transparency. 4. **Data science.** A data scientist asks different questions entirely. How much spam currently gets through? What does it cost in support tickets? Did the new filter actually reduce complaints, or did complaints fall for another reason? The answer might be an experiment, not a model. Notice that the first three build the filter, whereas the fourth decides whether the filter was worth building. ## When do you need deep learning, and when is classical ML enough? Deep learning is not automatically better. It excels on unstructured data such as images, audio and text. For the rows-and-columns data most businesses hold, however, classical methods often win. A 2022 benchmark by Grinsztajn, Oyallon and Varoquaux, [Why do tree-based models still outperform deep learning on tabular data?](https://arxiv.org/abs/2207.08815), found that tree-based models remained state of the art on medium-sized tabular datasets of around 10,000 samples. | Your data or constraint | Usual starting point | Why | |---|---|---| | Tables of customers, transactions or sensors | Regression or tree ensembles | Strong accuracy, fast to train, easier to explain | | Images, audio or video | Deep learning | Learns features that are impractical to hand-engineer | | Free text at scale | Deep learning, often a pre-trained model | Captures meaning beyond keyword counts | | Small dataset (hundreds of rows) | Simple models or rules | Too little data for a deep network to learn reliably | | Decisions that must be explained to regulators | Simpler, interpretable models | Easier to justify each prediction | | A "should we do this?" question | Data science methods, such as an experiment | The goal is a decision, not a prediction | ## Is data science part of AI? No, and this is the relationship people get wrong most often. Data science overlaps AI but is not contained by it. Experiment design, causal inference and descriptive analysis — a large share of the job — involve no machine learning whatsoever. The official job description makes the point too. The US Bureau of Labor Statistics lists five core duties for [data scientists](https://www.bls.gov/ooh/math/data-scientists.htm), and only one of them concerns building models. The rest cover finding data, analysing it, visualising findings and making recommendations. ## Which should you learn first? The sequencing mistake that costs the most time is starting with deep learning because it is the most exciting. Deep learning sits on top of machine learning, which in turn sits on top of statistics and programming. Skipping levels produces someone who can fine-tune a model but cannot tell you whether the result beats the baseline — exactly the candidate every interviewer is screening out. A sensible order looks like this: - statistics and Python, with SQL alongside - classical machine learning, including honest validation - deep learning, once a project genuinely needs unstructured data - applied AI, such as retrieval and evaluation, when you build on foundation models A structured program is worth paying for when it enforces that sequence rather than letting you jump to the interesting part. Deep learning research is the exception: it mostly happens in labs and usually requires a postgraduate background, so the path runs through academia far more often than through a course. ## Related reading To turn the concepts into a job decision, compare the roles in [Data Science vs Machine Learning vs AI](https://www.1stepgrow.com/articles/data-science-vs-machine-learning-vs-ai), including a three-day test for choosing between them. For the narrower decision, see [Data Analyst vs Data Scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist). For structured paths, we have roadmaps for [data science](https://www.1stepgrow.com/articles/data-scientist-roadmap-2026) and [AI engineering](https://www.1stepgrow.com/articles/ai-engineer-roadmap-2026), and a guide to [advanced data science and AI courses](https://www.1stepgrow.com/articles/advanced-data-science-ai-course). ## Frequently asked questions ### Is deep learning the same as AI? No. AI is the umbrella field, machine learning is one approach within it, and deep learning is one family of techniques within machine learning. Almost all recent progress people call 'AI' does come from deep learning, which is why the terms get conflated — but plenty of production AI systems use no deep learning at all. ### Is data science a subset of AI? No, and this is the relationship people get wrong most often. Data science overlaps AI but is not contained by it. A large part of data science — experiment design, causal inference, descriptive analysis — involves no machine learning whatsoever. ### Which should I learn first? Statistics and programming, before any of the four. After that, pick based on the working week you want rather than the salary table: data science if you like questions and persuasion, ML if you like systems, applied AI if you like shipping products fast. ### Do I need deep learning for a data science job? Usually not. Most data science work is regression, tree ensembles and experiment design. Deep learning matters for unstructured data — images, audio, text — and is over-represented in course syllabi relative to how often it appears in the job. ### Are these titles used consistently by employers? No, and you should assume they are not. Companies advertise 'Data Scientist' roles that are pure ML engineering and 'AI Engineer' roles that are mostly dashboards. Always read the responsibilities and the tech stack rather than the title. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/ai-vs-machine-learning-vs-deep-learning/). Cite with the title, "1stepGrow" and a link._ --- # Advanced Data Analytics Course: Choosing One Beyond Dashboards > Pick an advanced data analytics course that goes past dashboards: five options compared on price and predictive depth, plus the four topics a syllabus needs. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Aug 24, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 8 min - **Canonical URL:** https://www.1stepgrow.com/articles/advanced-data-analytics-course/ ## Key takeaways - Predictive analytics is the step that moves an analyst from reporting what happened to forecasting what will. - Most analytics courses stop at dashboards. Check whether regression, time series and validation are actually in the syllabus. - SQL depth matters more than any modelling technique on this list — it is what interviews test first. - A random train-test split on time series leaks future information; Forecasting: Principles and Practice recommends evaluating on a rolling forecasting origin instead. - Microsoft's PL-300 Power BI Data Analyst certification renews every 12 months through a free online assessment, and its exam price depends on the country. A good advanced data analytics course takes you past dashboards and into predictive work: regression you can defend, time series forecasting, honest validation, and enough SQL to survive a live interview round. Few programmes cover all four. Many stop at visualisation, because that is where a syllabus is easiest to build, so judge each one by its syllabus rather than its title. This guide is for analysts who have hit a specific ceiling: your SQL is good, your dashboards are clean, stakeholders trust your numbers, and you are still describing the past. It ranks five options, compares them on price, duration and predictive depth, lists the four topics a real syllabus contains, and says who should not buy an advanced course yet. ### 1. Learnbay — Data Analytics Certification Program Analytics through to predictive modelling, built for people holding down a job. - **Provider:** Learnbay - **Price:** On request (EMI available) - **Duration:** 9 months - **Level:** Beginner to intermediate - **1stepGrow score:** 4.5/5 - **Best for:** Analysts and non-tech professionals moving into predictive work - **Website:** https://www.learnbay.co/datascience/data-analytics-certification-course **Pros** - 275 hours covering live lectures, hands-on practice and interview preparation - Foundation programming classes included for non-programmers - IBM-certified capstone projects rather than tutorial exercises - Subscription access to repeat batches for 2-3 years — useful when work gets busy **Cons** - Nine months is a genuine commitment alongside a job - Pricing requires an enquiry **Verdict:** The most complete option here for someone who needs the full arc from SQL through to forecasting, with support attached. The repeat-batch access is a genuinely useful feature for working professionals who lose a month to a deadline. ### 2. Google Data Analytics Professional Certificate The best-value entry credential in analytics, with a hard ceiling. - **Provider:** Coursera - **Price:** $49/month - **Duration:** 3-6 months - **Level:** Beginner - **1stepGrow score:** 4.4/5 - **Best for:** Getting a first analyst role quickly and cheaply - **Website:** https://www.coursera.org/professional-certificates/google-data-analytics **Pros** - Well-structured and genuinely beginner-friendly - Recognised by a large employer consortium - Cheap relative to everything else **Cons** - Stops well short of predictive analytics - Very widely held — a floor, not a differentiator **Verdict:** Excellent first step. It will not teach you forecasting, so treat it as stage one of two. ### 3. Time Series Forecasting & Applied Regression The specific technical gap most analytics programs leave open. - **Provider:** University courses on Coursera / edX - **Price:** $49-300 - **Duration:** 6-10 weeks - **Level:** Intermediate - **1stepGrow score:** 4.3/5 - **Best for:** Analysts who already have SQL and want the predictive layer **Pros** - Directly targets the skill that separates analyst bands - Short and focused - Auditable free on several platforms **Cons** - Assumes statistics comfort - No career support **Verdict:** The highest return per hour on this list if you are already employed as an analyst. ### 4. Microsoft Power BI Data Analyst (PL-300) A certification, not a course — and a useful forcing function in Microsoft shops. - **Provider:** Microsoft - **Price:** $165 exam - **Duration:** 6-8 weeks prep - **Level:** Intermediate - **1stepGrow score:** 4/5 - **Best for:** Analysts in enterprise Microsoft environments - **Website:** https://learn.microsoft.com/en-us/credentials/certifications/data-analyst-associate/ **Pros** - Free, high-quality official learning path - Frequently tied to internal role banding **Cons** - Tool-specific — limited transfer - Barely touches predictive techniques **Verdict:** Worth it inside a Microsoft organisation, close to irrelevant outside one. ### 5. SQL Practice Platforms Not a course. Possibly the highest-value hours you will spend. - **Provider:** StrataScratch / DataLemur / LeetCode SQL - **Price:** Free-$100/year - **Duration:** Ongoing - **Level:** All levels - **1stepGrow score:** 4.5/5 - **Best for:** Everyone targeting an analytics role **Pros** - Directly targets the round most candidates fail - Cheap, immediate feedback **Cons** - No structure or credential **Verdict:** Do this alongside whatever else you pick. SQL is what gets tested first. ## How do the five options compare? The table below condenses the list. As of September 2026 these are the prices and durations shown above; fees change often, so confirm on the provider's own page before you pay. | Option | Format | Price | Duration | Predictive depth | Best for | |---|---|---|---|---|---| | Learnbay Data Analytics Certification Program | Live cohort with projects | On request | 9 months | Full arc, SQL to forecasting | Working analysts who need structure | | Google Data Analytics Professional Certificate | Self-paced | $49/month | 3-6 months | Stops before forecasting | A first analyst role | | Time series and applied regression courses | Self-paced university modules | $49-300 | 6-10 weeks | Narrow but deep | Analysts who already have SQL | | Microsoft PL-300 (Power BI) | Exam with free learning path | $165 exam (varies by country) | 6-8 weeks prep | Minimal | Microsoft-based teams | | SQL practice platforms | Practice, no teaching | Free-$100/year | Ongoing | None | Everyone, in parallel | Read the "predictive depth" column first. A cheap option that stops at dashboards is still the wrong course if forecasting is the gap you are trying to close. ## What should an advanced data analytics course cover? Read past the module titles. A program that genuinely teaches predictive analytics will cover four things, and most stop after the first. **Regression done properly.** Not just fitting a line — assumptions, residual analysis, multicollinearity, and knowing when the model is lying to you. **Time series.** Seasonality, trend, stationarity and why a naive train-test split leaks future information into your past. This is the single most commonly botched topic in analytics teaching. The free textbook *Forecasting: Principles and Practice* explains the fix, [evaluation on a rolling forecasting origin](https://otexts.com/fpp3/tscv.html), where every training set contains only observations that came before the test point. **Validation discipline.** Cross-validation, holdout design, and leakage detection. scikit-learn's own guide to [common pitfalls](https://scikit-learn.org/stable/common_pitfalls.html) defines data leakage as using information that would not be available at prediction time. If a course never mentions data leakage, its graduates will build models that look excellent in testing and fail in production. **Translating a forecast into a decision.** A prediction with no threshold and no action attached is a number, not analytics. The best programs make you state what someone should do differently. ## Live cohort, self-paced or exam prep: which format fits you? The format of an advanced data analytics course matters as much as its content, because an unfinished course teaches nothing. Match the delivery to how you actually work: - **Live cohort programmes** suit people who need deadlines and feedback on their code. They cost more and take longer, but the fixed schedule is the product. - **Self-paced certificates** suit disciplined learners on a budget. However, nobody checks whether your model is valid, so you have to supply that rigour yourself. - **Short university modules** suit working analysts with one clear gap, such as time series forecasting. They assume statistics comfort, so they are a poor first course. - **Certification exams** such as PL-300 suit people whose employer bands roles by credential. Microsoft states that the [Power BI Data Analyst certification](https://learn.microsoft.com/en-us/credentials/certifications/data-analyst-associate/) renews every 12 months through a free online assessment, and that the exam price depends on the country where you sit it. ## What does an analyst role actually require? Before you pay for depth, check what the job asks for. The US Department of Labor's O*NET profile for [Business Intelligence Analysts](https://www.onetonline.org/link/summary/15-2051.01) lists report generation, dashboard maintenance and trend analysis as core tasks. It also names SQL, Python, R, Power BI and Tableau among the technology skills. In other words, predictive modelling sits on top of reporting work rather than replacing it. That has a practical consequence. A course that teaches forecasting but skips data preparation leaves you unable to do the everyday half of the job. ## Who should skip an advanced course for now? An advanced programme is the wrong purchase in three situations: 1. **You cannot yet write a multi-table join.** Start with SQL and a beginner certificate; our [data analytics with Python guide](https://www.1stepgrow.com/data-analytics-course-with-python-online) covers that level. 2. **Your role has no forecasting problem to practise on.** Without real data to apply it to, the material fades within months. 3. **You are really aiming for data science.** In that case, compare the two paths first in [Data Analyst vs Data Scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist). ## Why should SQL practice run alongside any course? Because the first technical round for analyst roles tests SQL, and modelling courses rarely drill it hard enough. Every course on this list will teach you modelling. Almost none will make you good enough at SQL to pass the first interview round. That round is where analytics hiring actually filters. Not because the questions are exotic — window functions, CTEs, a tricky join — but because candidates spend six months on modelling and six days on SQL, then have to write a query while a stranger watches. If window functions are new to you, the [PostgreSQL tutorial on window functions](https://www.postgresql.org/docs/current/tutorial-window.html) is a clear, free starting point. Whatever program you choose, run SQL practice in parallel from week one. It is the highest-return hour in this entire guide. ## How do you get your employer to pay for it? Tie the course fee to a forecasting problem your organisation already pays for. Predictive analytics is one of the easiest upskills to build a business case for, and analysts consistently under-ask. The argument writes itself: better demand forecasting reduces inventory cost, churn prediction protects revenue, staffing forecasts cut overtime. Pick the one your organisation visibly struggles with, estimate the cost of getting it wrong today, and put the course fee next to that number. That framing succeeds far more often than "I would like a training budget", and it also tells you which specialisation to pick. ## Related reading For the career decision underneath this one, see [Data Analyst vs Data Scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist). For Python specifically, our [data analytics with Python guide](https://www.1stepgrow.com/data-analytics-course-with-python-online) covers the tooling route, and the [data scientist roadmap](https://www.1stepgrow.com/articles/data-scientist-roadmap-2026) shows where forecasting fits in a longer path. ## Frequently asked questions ### What is the difference between data analytics and predictive analytics? Analytics describes what happened and why. Predictive analytics estimates what will happen next, using regression, time series or classification. The second requires statistical validation skills the first does not, which is exactly why it commands higher pay. ### Do I need machine learning for predictive analytics? You need applied modelling — regression, tree ensembles, time series — plus the discipline to validate honestly. You do not need deep learning, and courses that lead with neural networks for tabular forecasting are teaching you the wrong tool. ### How much SQL do I actually need? More than most course syllabi suggest. Window functions, CTEs and multi-table joins should be comfortable, because a live SQL round is where a large share of otherwise strong analytics candidates are eliminated. ### Is predictive analytics being automated? AutoML has automated the model-fitting step, which was never the hard part. Framing the question, choosing the target variable, spotting leakage and deciding whether a forecast is trustworthy are all still human work — and are what these courses should be teaching. ### Will my employer pay for this? Frequently, and analysts under-ask. Predictive capability has a direct, arguable business case: better forecasting reduces inventory, churn or staffing cost. Make that argument rather than requesting a training budget in the abstract. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/advanced-data-analytics-course/). Cite with the title, "1stepGrow" and a link._ --- # Full Stack AI Course: When Breadth Pays Off and When It Does Not > Is a full stack AI course worth it? A 20-job-posting test for breadth versus depth, the layer most syllabi rush, and five programmes compared on price. - **Author:** Althaf Ashraf — AI Systems Engineer, Tata Consultancy Services (https://www.1stepgrow.com/authors/althaf-ashraf/) - **Published:** Aug 23, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Artificial Intelligence · **Format:** Listicle · **Read time:** 8 min - **Canonical URL:** https://www.1stepgrow.com/full-stack-artificial-intelligence-and-data-science-course/ ## Key takeaways - Full stack AI means owning the path from raw data to deployed system — genuinely valuable, but only if each layer is taught to sufficient depth. - The layer most programs skimp on is deployment and evaluation, which is precisely the layer employers cannot hire for. - Breadth is right for startups and small teams; large organisations hire specialists. - If a syllabus covers twelve tools in twelve weeks, you are buying a tour, not a skill. A full stack AI course is worth it when it has enough months to teach every layer properly — data pipelines, modelling, deployment, evaluation and generative AI — and when your target employers want one person to own the whole system. Otherwise, a specialist programme usually serves you better, because breadth without depth is hard to hire. "Full stack" sounds like more for your money: data engineering, modelling, deployment and generative AI in one programme. The risk is paying for all four and finishing shallow in each, which is a harder profile to hire than a specialist. That shape suits startups and small teams where one person owns the pipeline end to end. This page is for data scientists, engineers and career switchers comparing all-in-one programmes against focused ones. It ranks five options, shows what each layer of a serious syllabus should cover, and gives a simple job-posting test that settles the breadth-versus-depth question for your own target roles. ### 1. Learnbay — Forward Deployed AI Engineer Program Weighted toward the deployment end that most full stack programs treat as an afterthought. - **Provider:** Learnbay - **Price:** On request - **Duration:** 12 months - **Level:** Intermediate - **1stepGrow score:** 4.5/5 - **Best for:** Engineers who want to own AI systems in real customer environments - **Website:** https://www.learnbay.co/artificial-intelligence/forward-deployed-engineer-course **Pros** - Focuses on deploying into live customer environments, not just training models - Covers evaluation, cost and reliability — the production concerns that break demos - FDE-specific interview preparation, which almost nobody else offers - Job referrals through partner companies **Cons** - Twelve months, and it assumes real programming ability - Narrower than a generalist data science program by design **Verdict:** The strongest option here precisely because it refuses to be a tour. It goes deep on the layer — deployment and evaluation — where the hiring demand actually is. ### 2. Learnbay — Advanced Data Science & GenAI Master Program The broader sibling: fundamentals, modelling and generative AI across thirteen months. - **Provider:** Learnbay - **Price:** On request (EMI available) - **Duration:** 13 months - **Level:** Intermediate - **1stepGrow score:** 4.6/5 - **Best for:** People who want genuine breadth with time to absorb it - **Website:** https://www.learnbay.co/datascience/data-science-and-ai-master-certification-program **Pros** - Long enough that breadth does not become shallowness - Four capstones with IBM project certification - Domain electives so projects match your industry **Cons** - Thirteen months is the longest commitment on this list **Verdict:** Choose this over the FDE program if you want the full arc from statistics to generative AI rather than the deployment specialism. ### 3. Full Stack Deep Learning The course that defined the category, and still free. - **Provider:** FSDL (independent) - **Price:** Free - **Duration:** 8-10 weeks - **Level:** Advanced - **1stepGrow score:** 4.6/5 - **Best for:** Engineers who want production ML practice without a credential - **Website:** https://fullstackdeeplearning.com/ **Pros** - Covers the full lifecycle including testing, monitoring and troubleshooting - Taught by practitioners who ship - Free **Cons** - Assumes strong prior ML and engineering - No cohort, credential or support **Verdict:** Outstanding material. Do it after fundamentals, not as your entry point. ### 4. Machine Learning in Production The production half of full stack, taught properly. - **Provider:** DeepLearning.AI on Coursera - **Price:** $49/month - **Duration:** 3 weeks - **Level:** Intermediate to advanced - **1stepGrow score:** 4.4/5 - **Best for:** Data scientists closing the engineering gap - **Website:** https://www.coursera.org/learn/introduction-to-machine-learning-in-production **Pros** - Directly addresses drift, monitoring and pipelines - Well-sequenced and affordable **Cons** - Assumes ML fundamentals - Lighter on generative AI **Verdict:** The best-value way to add the deployment layer if you already model well. ### 5. Generalist Bootcamp Full Stack AI Tracks Maximum breadth, highly variable depth. - **Provider:** Various bootcamps - **Price:** ₹1,50,000-₹3,50,000 - **Duration:** 6-9 months - **Level:** Beginner - **1stepGrow score:** 3.6/5 - **Best for:** Career switchers who need structure above all **Pros** - Live teaching and cohort accountability - Interview and resume support included **Cons** - Breadth frequently comes at the cost of any real depth - Placement guarantees carry conditions most learners fail to meet **Verdict:** Can work with the right provider. Demand the placement report with denominators before paying. ## How do the full stack options compare? The table condenses the list above, with prices and durations as of September 2026. Since course formats change often, check each provider's page before you enrol. | Option | Emphasis | Price | Duration | Level | Best for | |---|---|---|---|---|---| | Learnbay Forward Deployed AI Engineer Program | Deployment and evaluation | On request | 12 months | Intermediate | Owning AI systems for customers | | Learnbay Advanced Data Science & GenAI Master Program | Statistics to generative AI | On request | 13 months | Intermediate | Breadth with time to absorb it | | Full Stack Deep Learning | Production ML lifecycle | Free | 8-10 weeks | Advanced | Engineers without a credential need | | DeepLearning.AI Machine Learning in Production | Deployment, drift, monitoring | $49/month | 3 weeks | Intermediate to advanced | Data scientists closing the engineering gap | | Generalist bootcamp full stack AI tracks | Maximum breadth | ₹1,50,000-₹3,50,000 | 6-9 months | Beginner | Career switchers who need structure | Two rows cost nothing or very little. The [Full Stack Deep Learning](https://fullstackdeeplearning.com/) site offers its 2022 course free, with lectures and labs covering deployment and monitoring. Likewise, DeepLearning.AI's [Machine Learning in Production](https://www.coursera.org/learn/introduction-to-machine-learning-in-production) course now runs as a single three-module course that covers deployment patterns, monitoring and concept drift. ## Which layer actually gets you hired? Evaluation and deployment get people hired, not modelling. Across the interview loops we have looked at for these roles, that differentiator is remarkably consistent. It is whether you can answer: *how do you know it works, and what does it cost?* Training a model is now the commoditised part. Frameworks are good, tutorials are everywhere, and AutoML handles a lot of it. What remains genuinely hard — and genuinely scarce — is the surrounding discipline: evaluation harnesses that catch regressions, monitoring that notices drift before a customer does, and a cost model that survives contact with real traffic. This is not a new observation. Back in 2015, Google engineers argued in [Hidden Technical Debt in Machine Learning Systems](https://papers.nips.cc/paper/2015/hash/86df7dcfd896fcaf2674f757a2463eba-Abstract.html) that quick ML wins carry large ongoing maintenance costs from data dependencies, feedback loops and configuration. Even so, that is the layer most full stack curricula compress into the final two weeks. It is also the layer employers say they cannot hire for. ## What should a full stack AI course cover at each layer? Use Google Cloud's description of [MLOps maturity levels](https://docs.cloud.google.com/architecture/mlops-continuous-delivery-and-automation-pipelines-in-machine-learning) as a yardstick. That guide notes that only a small fraction of a real-world ML system is the ML code itself. It then describes a progression from manual, notebook-driven work (level 0) to automated training pipelines (level 1) and full CI/CD for those pipelines (level 2). A serious syllabus should take you at least to level 1. In practice, that means: - **Data:** building a reliable ingestion pipeline and validating data before training. - **Modelling:** training, tuning and comparing models against a simple baseline. - **Deployment:** serving a model behind an API, with a rollback plan. - **Monitoring and evaluation:** detecting drift, tracking cost and catching regressions after release. - **Generative AI:** building on the evaluation skills above rather than replacing them. If the course stops at a notebook, it is teaching level 0 and calling it full stack. ## Should you choose breadth or depth? Let the job postings decide, not the syllabus. Open twenty postings for the roles you want. If they read "own the ML pipeline end to end, small team, wear many hats", breadth is correct and a full stack program is a good fit. By contrast, if they read "you will work with our data platform team and our MLOps team", you are being hired as a specialist, so a generalist course will leave you competing badly in every round. Regional and company-size patterns are strong here. Startups and services companies lean generalist. Large product organisations, on the other hand, lean specialist almost without exception. ## Who should skip a full stack AI course? A broad programme is the wrong purchase if: 1. **You already model well.** Add the deployment layer with a focused course; the [MLOps career guide](https://www.1stepgrow.com/articles/mlops-engineer-career-guide) explains what that role involves. 2. **You cannot yet program.** Start with Python fundamentals first, because every layer here assumes it. 3. **Your target employers hire specialists.** Then follow a specialist path such as the [AI Engineer roadmap](https://www.1stepgrow.com/articles/ai-engineer-roadmap-2026) instead. ## What should you check before paying for a full stack AI course? Check what you will ship and how it is assessed, because module lists all look alike. **Count deliverables, not modules.** Four systems you deployed and can discuss for forty minutes beats a forty-module syllabus every time. **Ask what the capstone output is.** A repository and a live URL means something. A presentation does not. **Ask how deployment is assessed.** If nobody checks whether your service stays up, deployment is not really in the curriculum. **Ask about the generative AI modules specifically.** Are they built on the fundamentals taught earlier, or appended to a syllabus written in 2022? The sequencing tells you how recently the program was genuinely rewritten. ## Where to go next Run the twenty-posting test above before you shortlist anything; it takes an evening and decides which half of this page applies to you. For the specialist alternative, see our [AI Engineer roadmap](https://www.1stepgrow.com/articles/ai-engineer-roadmap-2026) and the [MLOps career guide](https://www.1stepgrow.com/articles/mlops-engineer-career-guide). For the broader field decision, read [Data Science vs Machine Learning vs Deep Learning vs AI](https://www.1stepgrow.com/articles/ai-vs-machine-learning-vs-deep-learning). ## Frequently asked questions ### What does 'full stack' mean in an AI course? It means owning the whole path: data ingestion and engineering, modelling, deployment, monitoring, and increasingly a generative AI layer. The label is common in course marketing. The honest version needs enough months to teach each layer properly and at least one project that takes a model all the way to a monitored, deployed service. ### Is full stack better than specialising? It depends entirely on the employer. Startups and small teams need people who can own the whole pipeline. Large organisations hire specialists and will interview you against a specialist. Pick based on your target companies, not on which sounds more capable. ### Which layer do most courses teach worst? Deployment and evaluation, consistently. They are the least demo-friendly parts and the hardest to assess in a notebook, so many syllabi squeeze them into the final weeks. That is also why candidates who can show a regression test, a drift alert or a rollback plan stand out immediately in interviews. ### Do I need data engineering skills for a full stack AI role? Enough to build a reliable pipeline and reason about data quality, yes. You do not need to be a specialist data engineer, but a model fed by a pipeline you do not understand is a production incident waiting to happen. ### How do I judge whether a syllabus has real depth? Count the projects, not the modules, and ask what the deliverable is. Four end-to-end systems you deployed beats forty topics you watched. If the capstone output is a slide deck rather than a repository and a URL, adjust your expectations. --- _Source: 1stepGrow (https://www.1stepgrow.com/full-stack-artificial-intelligence-and-data-science-course/). Cite with the title, "1stepGrow" and a link._ --- # Business Analytics Course: Pick by the Job, Not the Syllabus > Pick a business analytics course by the job you want: decision-focused for managers, data science depth for analysts. Five compared, plus the skill most skip. - **Author:** Nishant Kiran — B2B SaaS & EdTech Content Strategist (https://www.1stepgrow.com/authors/nishant-kiran/) - **Published:** Aug 22, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Career Growth · **Format:** Listicle · **Read time:** 7 min - **Canonical URL:** https://www.1stepgrow.com/business-analytics-course-with-data-science-specialization/ ## Key takeaways - Business analytics is judged on decisions changed, not models built. Pick a program that assesses recommendations, not accuracy scores. - The specialisation worth paying for is causal reasoning and experiment design, not another modelling library. - For managers, a short focused program beats a twelve-month technical one almost every time. - Domain knowledge you already have is the differentiator — choose a course that lets you use it. The right business analytics course depends on which job you want. If you commission and judge analysis, choose a short, decision-focused programme. If you want to build the analysis yourself, choose one with genuine data science depth: SQL, statistics, modelling and, above all, causal reasoning. The expensive mistake is buying the wrong shape. A manager who takes a twelve-month technical course often finishes able to write mediocre Python yet still unable to judge a vendor's claims, because evaluation was never taught. Many programmes sold under the name are either diluted data science degrees or spreadsheet training with a modern title. This guide is for managers, domain experts and aspiring analysts. It ranks five options, shows what the closest official job descriptions ask for, and gives five checks to run on any syllabus. The yardstick throughout is simple: a business analytics professional is judged on decisions changed, not models built. ### 1. Learnbay — GenAI & Agentic AI for Managers & Leaders Written for people who decide about AI rather than build it — four months, not twelve. - **Provider:** Learnbay - **Price:** On request - **Duration:** 4 months - **Level:** Intermediate - **1stepGrow score:** 4.4/5 - **Best for:** Managers and business leaders who must make AI and analytics calls - **Website:** https://www.learnbay.co/genai-agentic-ai-managers-leaders **Pros** - Aimed at evaluation, risk and adoption rather than implementation detail - Four months part-time is realistic for someone running a team - IBM GenAI certification - Does not pretend you need to become an engineer **Cons** - Deliberately not a technical build course — you will not come out writing pipelines - Pricing requires an enquiry **Verdict:** The right shape for the audience. Most business analytics programs fail managers by teaching them a diluted version of the engineer's curriculum; this one teaches the decisions they actually have to make. ### 2. Learnbay — Data Analytics Certification Program The hands-on route if you want to build the analysis yourself. - **Provider:** Learnbay - **Price:** On request (EMI available) - **Duration:** 9 months - **Level:** Beginner to intermediate - **1stepGrow score:** 4.5/5 - **Best for:** Business professionals who want to do the analysis, not commission it - **Website:** https://www.learnbay.co/datascience/data-analytics-certification-course **Pros** - Foundation classes for non-programmers included - IBM-certified capstone projects - Weekday and weekend batches **Cons** - Nine months, and genuinely hands-on **Verdict:** Pick this over the managers program if you want to build the models rather than evaluate them. ### 3. Business Analytics Specialisation Business-school framing, strong on decisions and weak on tooling. - **Provider:** Coursera (Wharton) - **Price:** $49/month (first module free to preview) - **Duration:** 4-6 months - **Level:** Beginner - **1stepGrow score:** 4.3/5 - **Best for:** Managers who want the framework without the code - **Website:** https://www.coursera.org/specializations/business-analytics **Pros** - Excellent on framing problems and interpreting results - Covers marketing, people and operations analytics - Auditable free **Cons** - Very light on hands-on technical work - No placement or career support **Verdict:** Strong conceptual grounding, minimal practical capability. Good as a first or parallel course. ### 4. Executive PG in Business Analytics The credential play for internal promotion. - **Provider:** IIM / ISB via EdTech partners - **Price:** ₹2,00,000-₹4,00,000 - **Duration:** 10-12 months - **Level:** Intermediate - **1stepGrow score:** 3.9/5 - **Best for:** Professionals who need institutional branding for a step up **Pros** - Brand clears HR filters and internal banding committees - Peer cohort of senior professionals has real network value **Cons** - Delivery frequently by the EdTech partner rather than institute faculty - Content often lags practice **Verdict:** You are buying the brand and the network. Judge it on those terms and it can be worth it. ### 5. Experimentation & Causal Inference courses The actual differentiator, and the least-taught topic in business analytics. - **Provider:** Various (Coursera, edX, independent) - **Price:** Free-$300 - **Duration:** 6-10 weeks - **Level:** Intermediate - **1stepGrow score:** 4.4/5 - **Best for:** Anyone who has to prove something caused something **Pros** - Directly targets the skill senior analytics roles are hired for - Transfers across every industry **Cons** - Requires statistics comfort - Rarely bundled into business analytics programs **Verdict:** The most under-rated addition to any business analytics education. Add it to whatever else you choose. ## How do the five programmes compare? This table condenses the list above, with prices and durations as of September 2026. Since fees and batch structures change, confirm the details on each provider's page. | Programme | Built for | Price | Duration | Hands-on depth | |---|---|---|---|---| | Learnbay GenAI & Agentic AI for Managers & Leaders | Commissioning role | On request | 4 months | Low by design | | Learnbay Data Analytics Certification Program | Building role | On request | 9 months | High | | Wharton Business Analytics Specialisation (Coursera) | Either, as grounding | $49/month (first module free to preview) | 4-6 months | Low | | Executive PG in Business Analytics (IIM / ISB via partners) | Credential seekers | ₹2,00,000-₹4,00,000 | 10-12 months | Medium | | Experimentation & causal inference courses | Anyone proving cause | Free-$300 | 6-10 weeks | Medium to high | A note on the Wharton row: Coursera's [Business Analytics Specialization page](https://www.coursera.org/specializations/business-analytics) estimates about two months at 10 hours a week, so the 4-6 month range assumes a lighter weekly load. The page also lists its five parts — customer, operations, people and accounting analytics, plus a capstone — and says no prior experience is required. Since August 2025, Coursera has [replaced free auditing](https://blog.coursera.org/introducing-courseras-new-course-preview-experience/) with a free preview of the first module of nearly every course, so older "audit free" claims no longer hold. ## Which business analytics job are you training for? Decide whether you will commission analysis or build it. This fork determines everything, and most people skip it. **The commissioning role.** You lead a team or a function. You need to frame questions well, tell a good analysis from a bad one, know when a result is being oversold, and decide what to do about it. Nobody expects you to write the code. **The building role.** You do the analysis yourself. That means SQL, a modelling toolkit, and the discipline to validate honestly. These need different programmes, and the mismatch cuts both ways. The manager on a technical course learns tooling instead of evaluation, while a would-be analyst who takes a conceptual course finishes unable to produce anything. ## What does a business analytics role actually involve? It involves turning data into a recommendation someone acts on, and official job descriptions make that decision focus explicit. The US Bureau of Labor Statistics describes [operations research analysts](https://www.bls.gov/ooh/math/operations-research-analysts.htm), one of the closest official matches, as people who gather data, build quantitative models and then present recommendations to decision-makers on resources, scheduling, supply chains and pricing. It lists median US pay of $88,940 in May 2025. Similarly, the O*NET profile for [Business Intelligence Analysts](https://www.onetonline.org/link/summary/15-2051.01) centres on reports and trend analysis for executives, and names SQL, Python, Power BI and Tableau among the tools. So a business analytics course for the building role should teach both the tooling and the recommendation that follows it. ## How should you evaluate a business analytics course before paying? Use these checks on any syllabus, whichever role you picked: 1. **Look at how projects are assessed.** A good course grades the recommendation and its reasoning, not only the accuracy of a model. 2. **Search the syllabus for causal inference or experiment design.** If neither appears, plan to add a separate course. 3. **Ask whether you can use your own industry's data.** Domain projects are far more useful than a generic public dataset. 4. **Check who teaches.** For institute-branded programmes, confirm whether faculty or the EdTech partner delivers the sessions. 5. **Match the length to your role.** Four to six months suits a decision-focused programme; nine to twelve suits a hands-on one. ## Which skill matters most in a business analytics course? Causal inference and experiment design. If you take one recommendation from this page, make it that one. Business analytics questions are almost always causal. Did the campaign work? Would this pricing change increase revenue? Does the training program reduce attrition? These are not prediction problems, and treating them as prediction problems is the most common serious error in corporate analytics. Very few business analytics programmes teach this properly. The ones that do produce people whose recommendations survive scrutiny, which is the entire job. If you want a rigorous free reference, Miguel Hernán and James Robins make their textbook [*Causal Inference: What If*](https://miguelhernan.org/whatifbook) available online. ## Who should skip a business analytics course? A course is not always the best next step. Skip it, at least for now, if: - **You already run experiments competently.** In that case, a focused causal inference or forecasting module adds more than a broad programme. - **You want to become a data scientist.** A business analytics course is the wrong shape; read [Data Analyst vs Data Scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist) first. - **Your organisation will not act on analysis.** Then no programme fixes the real problem, so raise it before spending money. ## Why does your existing domain knowledge matter so much? Because it is the input a course cannot supply. The people who get the most out of business analytics programmes are not the ones with the strongest maths; they are the ones who arrive with ten years in insurance, retail operations or hospital administration. That knowledge is the scarce input. Knowing which numbers are unreliable, which stakeholder will object and why, and what a plausible effect size looks like in your industry takes years to acquire and cannot be taught in a cohort. Choose a program that lets you bring your own domain into the projects. A capstone on your industry's actual problem is worth five on a public dataset — both for what you learn and for what you can show afterwards. ## Where to go next Write down whether you want to commission or build analysis, then run the five syllabus checks above on your top two options. For the adjacent technical route, see [Data Analyst vs Data Scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist) and our guide to [advanced data analytics with predictive analytics](https://www.1stepgrow.com/articles/advanced-data-analytics-course). Managers evaluating AI specifically may want the [AI manager specialisation guide](https://www.1stepgrow.com/data-science-specialization-course-ai-manager). ## Frequently asked questions ### What is the difference between business analytics and data science? Business analytics starts from a business decision and works backward to the data. Data science more often starts from the data and works toward what can be learned. In practice the overlap is large, but business analytics is judged on decisions changed and data science on the rigour of the finding. ### Do I need to learn coding for business analytics? For a management role, no: you need to interpret, question and commission analysis rather than write it. For a hands-on analyst role, yes: SQL at minimum, and Python increasingly, alongside a BI tool such as Power BI or Tableau. Decide which of the two jobs you actually want before choosing a programme, because the courses differ sharply. ### Is a business analytics certificate worth it for a manager? If it teaches evaluation and decision-making, yes — it is what stops you approving a project that cannot work. If it is a diluted version of an engineer's syllabus, you will finish it able to neither build nor properly evaluate, which is the worst of both. ### What actually gets a business analytics professional promoted? Being the person whose recommendations get acted on. That comes from framing questions well, being honest about uncertainty, and communicating in a way that survives a sceptical room. None of those are modelling skills. The technical skill that helps most is causal reasoning, because it lets you say with confidence whether a change actually caused a result. ### How long should a business analytics program take? Four to six months for a decision-focused programme, and nine to twelve for a hands-on one that teaches SQL, statistics and modelling. Anything longer should be justified by a credential you specifically need, such as an institute brand for an internal promotion. A short causal inference course of six to ten weeks can sit alongside either. --- _Source: 1stepGrow (https://www.1stepgrow.com/business-analytics-course-with-data-science-specialization/). Cite with the title, "1stepGrow" and a link._ --- # Data Science Course for Managers: What AI Leaders Actually Need > What a data science course for managers should teach AI leaders: the four decisions you will own, the literacy floor and five checks before you pay. - **Author:** Nishant Kiran — B2B SaaS & EdTech Content Strategist (https://www.1stepgrow.com/authors/nishant-kiran/) - **Published:** Aug 21, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Career Growth · **Format:** Guide · **Read time:** 7 min - **Canonical URL:** https://www.1stepgrow.com/data-science-specialization-course-ai-manager/ ## Key takeaways - A manager's job is evaluation, not implementation. Courses that teach a watered-down engineering syllabus fail at both. - The four decisions leaders repeatedly get wrong: what to build, how to measure it, what it costs at scale, and when to stop. - Four to six months part-time is the right size. Twelve-month technical programs are usually a poor fit for people running teams. - You need enough technical literacy to ask uncomfortable questions, not enough to write the code. A good data science course for managers teaches judgement, not implementation: how to decide what to build, how to measure it, what it costs at scale and when to stop. Four to six months part-time is enough, provided the syllabus weights evaluation and risk far above model architectures. Many "AI for managers" courses take an engineer's curriculum, remove the hard parts and sell the remainder to people who lead teams. The result is a manager who can describe a transformer at a dinner party but cannot tell whether the model their team shipped last quarter is any good. Google's own Rules of Machine Learning start somewhere else entirely: its first rule is not to be afraid to launch a product without machine learning. This guide is for team leads, product owners and executives who approve or oversee AI work. It sets out the four decisions you will actually make, the minimum technical literacy behind them, a governance framework you can borrow, and five checks for choosing a programme. ## How does a manager's curriculum differ from an engineer's? The difference is easiest to see side by side. Both roles touch the same systems, but they need different skills from them. | Topic | What an engineer learns | What a manager needs | |---|---|---| | Problem selection | How to frame a prediction task | Whether a model is needed at all | | Evaluation | How to compute metrics | Which metric defines success, agreed in advance | | Cost | How to optimise inference | Unit economics at ten times the pilot volume | | Failure modes | How to debug leakage or drift | How to spot the warning signs in a status report | | Governance | How to implement controls | Who owns the decision to stop | A course that fills the middle column and ignores the right-hand one is aimed at the wrong reader, however "executive" its title sounds. ## Which four decisions does an AI manager actually make? What to build, how to measure it, what it costs at scale and when to stop. Everything a leader needs from this education serves those four recurring decisions. ### Choosing the problem and the measure **What to build.** Which problems are genuinely suited to a model, and which are being forced into one because AI is in the strategy deck. The most valuable skill here is recognising when the answer is a rule, a lookup table or a process change — and saying so before six months are spent. Google's widely used [Rules of Machine Learning](https://developers.google.com/machine-learning/guides/rules-of-ml) opens with exactly this advice: its first rule is not to be afraid to launch a product without machine learning. **How to measure it.** What would count as success, defined before the work starts, in a number someone will be accountable for. This is the single most common gap. A project without a pre-agreed success metric cannot fail, which sounds comfortable and is why bad projects run for years. The same Google guide's second rule is to design and implement metrics first. ### Counting the cost and knowing when to stop **What it costs at scale.** Inference cost, latency, human review, retraining, monitoring. Pilots hide all of this. The decision to go to production is a cost decision far more often than a technical one, and the people making it frequently have never seen the unit economics. The classic paper [Hidden Technical Debt in Machine Learning Systems](https://papers.nips.cc/paper/2015/hash/86df7dcfd896fcaf2674f757a2463eba-Abstract.html) (NeurIPS 2015) makes the same point: quick wins in ML carry maintenance costs that do not show up at the start. **When to stop.** Both stopping a project that is not working and stopping a model that has quietly degraded. Somebody has to own that call, and it is not the person who built it. None of the four requires you to write code. All four require enough literacy to ask uncomfortable questions and understand the answers. ## How much technical literacy does an AI manager need? Enough to read charts, evaluations and failure reports critically. That floor is lower than most managers fear but higher than most courses deliver. **Read a chart honestly.** Know what a confidence interval means, why an accuracy score can be worthless on imbalanced data, and why the y-axis starting at 80% is a choice. **Understand evaluation.** The difference between offline metrics and online outcomes, why a model that scores well in testing can fail in production, and what a holdout set is for. **Recognise the common failure modes.** Data leakage, drift, feedback loops where the model's own output becomes its next input, and the specific ways generative systems fail — confident wrongness, prompt injection, cost blowouts. The [OWASP Top 10 for LLM Applications](https://genai.owasp.org/llm-top-10/) is a useful checklist here. As of September 2026, its latest edition (2025) ranks prompt injection first and lists unbounded consumption, which includes runaway usage costs, as its tenth risk. **Ask about the data.** Where it came from, what it excludes, who is under-represented in it, and what would happen if the source changed. That list is learnable in four months part-time. It is also enough to make you a materially better decision-maker than most people currently approving AI budgets. ## Which frameworks help managers govern AI projects? You do not need to invent a governance process from scratch. The US National Institute of Standards and Technology published its [AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) on 26 January 2023, followed by a Generative AI Profile on 26 July 2024. The framework is voluntary, and it organises the work into four functions: - **Govern:** who is accountable, and which policies apply. - **Map:** the context, the intended use and who could be affected. - **Measure:** how risks and performance are assessed. - **Manage:** what you do about the risks you found, including stopping. A strong course will teach at least one framework like this, because it turns the four decisions above into a repeatable review rather than a matter of personal instinct. ## What should a data science course for managers include? Evaluation, cost and risk should dominate the syllabus, and assessment should be based on decisions. When you compare programmes, check these five points: 1. **Duration proportionate to the role.** Four to six months. If you are running a team, a twelve-month technical program is a commitment you will not sustain, and the parts you would skip are the parts you needed. 2. **Evaluation weighted heavily.** Scan the syllabus for how much time goes to measurement, risk and adoption versus model architectures. The ratio tells you who the course is really for. 3. **Case-based assessment.** You should be judged on decisions and their justification, not on whether your model hit an accuracy threshold. 4. **A cohort of peers.** Much of the value in manager-level programs is the other people in the room facing the same decisions. This is one context where the network argument is genuinely true rather than marketing. 5. **Current generative AI content.** Look for evaluation and risk of LLM systems, not just demonstrations of prompting. ## Who should skip a data science course for managers? This kind of programme is not for everyone. Choose a different route if: - **You want to build the models yourself.** A hands-on analytics or data science course fits better; our [business analytics guide](https://www.1stepgrow.com/business-analytics-course-with-data-science-specialization) compares both routes. - **Your organisation has no AI project for you to oversee.** Without one, the judgement never forms, so wait until you have a live project. - **You already review AI work regularly.** In that case, a short course on generative AI risk adds more than a full programme; see our list of [generative AI certifications](https://www.1stepgrow.com/articles/best-genai-certifications-2026). ## Why does one real project teach more than the course? Because judgement only forms when a decision has consequences. Whatever programme you choose, run one real project alongside it that you personally oversee from problem statement to production decision. Write the success metric down before the work starts. Insist on seeing the evaluation. Ask what it costs at ten times the volume. Then, at the end, write an honest note about what you got wrong. That document will teach you more than any module, and it is what makes the next decision better. ## Where to go next Before you enrol anywhere, write the success metric for one AI project you currently oversee; if you cannot, that is the first gap to close. For the technical grounding underneath these decisions, see [Data Science vs Machine Learning vs Deep Learning vs AI](https://www.1stepgrow.com/articles/ai-vs-machine-learning-vs-deep-learning). For the hands-on route, our [business analytics guide](https://www.1stepgrow.com/business-analytics-course-with-data-science-specialization) covers the build-it-yourself option. ## Frequently asked questions ### Do AI managers need to code? No, but you need to read technical work. Being able to follow a notebook, understand what an evaluation metric measures and spot when a chart has been chosen to flatter a result is essential. Writing production code is not. The test is whether you can ask a question about the evaluation that your team has to think about before answering. ### What is the biggest mistake managers make with AI projects? Approving a project with no definition of success. If nobody can state before the work starts what number would need to move and by how much, the project cannot fail visibly, and so it never gets stopped. Agree the metric, the threshold and a review date in writing before any build work begins. ### How technical should an AI manager course be? Technical enough to teach evaluation, cost modelling and failure modes, but not so technical that it becomes a diluted engineering curriculum. If the syllabus spends more time on model architectures than on measurement, risk and adoption, it is aimed at the wrong reader. Case-based assessment is a good sign; accuracy-threshold assignments are not. ### Is an AI manager role a real career step? Increasingly, yes. Organisations that have shipped a few AI projects discover they need someone who can judge them: set success metrics, weigh cost at scale and decide when to stop. That person rarely comes from a purely engineering or purely business background, which is why people who combine domain experience with evaluation literacy are well placed. ### How long does it take to become effective? Four to six months of part-time study plus one real project you oversee end to end. The project matters more than the course, because it is where the judgement forms. Write the success metric down first, insist on seeing the evaluation, and finish with an honest note on what you got wrong. --- _Source: 1stepGrow (https://www.1stepgrow.com/data-science-specialization-course-ai-manager/). Cite with the title, "1stepGrow" and a link._ --- # Data Science vs Machine Learning vs AI: Which Career Suits You? > Data science vs machine learning vs AI as careers: what each role does day to day, the skills and official pay data, and a three-day test to pick your path. - **Author:** Althaf Ashraf — AI Systems Engineer, Tata Consultancy Services (https://www.1stepgrow.com/authors/althaf-ashraf/) - **Published:** Aug 20, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Comparison · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/data-science-vs-machine-learning-vs-ai/ ## Key takeaways - Data science is a communication job with statistics attached. ML engineering is a software job with modelling attached. AI engineering is a product job with APIs attached. - The three roles have converged in job titles but diverged in day-to-day work — read the responsibilities, never the title. - ML engineering pays the most and has the highest barrier to entry because it needs both software and modelling skill. - Switching between the three later is far easier than people expect. Picking wrong costs you months, not years. Data science vs machine learning vs AI, as careers, comes down to three different jobs. Data scientists turn business questions into decisions. ML engineers build and run the systems that serve models. AI engineers build products on top of foundation models. Choose by the working week you want, not by the title or the median salary. Job boards have blurred the titles: one company will advertise a "Data Scientist" role that is pure ML engineering and an "AI Engineer" role that is mostly dashboard building. The underlying jobs still reward different temperaments, need different preparation and pay differently. Choosing by title is how people spend eighteen months preparing for the wrong interview. This guide is for students and career changers choosing a direction. It sets out each role's working week, the skills in learning order, what official pay data does and does not show, and a three-day test you can run yourself. If you first want the conceptual difference — how AI, machine learning and deep learning nest inside each other — read [AI vs machine learning vs deep learning](https://www.1stepgrow.com/articles/ai-vs-machine-learning-vs-deep-learning), then come back here. | Feature | Data Scientist | ML Engineer | AI Engineer | | --- | --- | --- | --- | | Core daily work | SQL, analysis, experiment design, stakeholder decks | Pipelines, model serving, monitoring, infrastructure | Prompting, retrieval, evaluation harnesses, API integration | | Primary language | Python + heavy SQL | Python + Go/Java, plus IaC | Python/TypeScript | | Maths depth needed | High — statistics, inference | Medium — applied, not theoretical | Low to medium | | Software engineering depth | Low to medium | High | Medium | | Entry difficulty from non-tech background | Moderate | Hard | Easiest of the three | | Median India salary (3-5 yrs) | ₹18 LPA | ₹26 LPA | ₹22 LPA | | Remote availability | Medium | High | High | | Automation risk over 5 years | Medium — routine reporting is going | Low | Medium — the tooling is commoditising | | Portfolio that gets interviews | An analysis that changed a decision | A deployed service with monitoring | A working product with real users | | Typical team size you sit in | 2-6 analysts/scientists | 3-10 engineers | 1-4, often embedded in product | **Verdict — ML Engineer:** On pure numbers — pay, demand, automation resistance and career optionality — ML engineering is the strongest of the three in 2026. But it is also the hardest to enter, because it requires you to be competent at two disciplines rather than very good at one. The right answer for you depends far more on which of the three working weeks you would actually enjoy than on which has the best median salary. ## Data science vs machine learning vs AI: what does each week look like? The fastest way to choose is to imagine the week, not the title. **A data scientist's week** is maybe forty percent SQL and analysis, thirty percent meetings and stakeholder conversation, twenty percent modelling and ten percent making slides. The hardest part of the job is usually not technical. Instead, it is establishing that a result is real, then persuading someone to act on it. If explaining a confidence interval to a sceptical VP sounds satisfying rather than exhausting, this is your role. **An ML engineer's week** looks like software engineering with a statistical component: pipelines, deployment, monitoring and the pager. The model itself is often the smallest part, while the real work is everything around it that must not break at 3am. Google researchers made the same point in [Hidden Technical Debt in Machine Learning Systems](https://papers.nips.cc/paper_files/paper/2015/hash/86df7dcfd896fcaf2674f757a2463eba-Abstract.html), which argues that real-world ML systems commonly incur massive ongoing maintenance costs. If debugging a data drift alert interests you more than presenting a finding, this is your role. **An AI engineer's week** is product engineering against a non-deterministic component: building retrieval, writing evaluations, controlling cost and latency, and shipping. It is the fastest feedback loop of the three and also the most volatile, because the tooling changes underneath you every quarter. If that sounds energising rather than exhausting, this is your role. ## Which skills does each career path need? All three share Python and a working grasp of statistics. After that, the priorities diverge sharply. | Skill | Data scientist | ML engineer | AI engineer | |---|---|---|---| | SQL | Essential, daily | Useful | Useful | | Statistics and experiment design | Essential | Applied level | Enough to design evaluations | | Software engineering (testing, CI, code review) | Helpful | Essential | Essential | | Cloud, containers and deployment | Rarely needed | Essential | Important | | Classical ML (regression, tree ensembles) | Essential | Essential | Helpful | | Foundation model APIs, retrieval, evaluation | Helpful | Helpful | Essential | | Stakeholder communication | Essential | Moderate | Important, with product teams | Read the table as a learning order. Whatever is "Essential" in your chosen column comes first, and "Helpful" skills can wait until a job asks for them. ## Why is evaluation the core AI engineering skill? AI engineering is not prompt writing. The job exists because model output is useful but unreliable, so someone has to measure it. The [2025 Stack Overflow Developer Survey](https://survey.stackoverflow.co/2025/ai) shows the tension clearly: 84% of respondents were using or planning to use AI tools in development, yet 46% actively distrusted the accuracy of those tools, against 33% who trusted it. The most common frustration, cited by 66% of respondents, was AI output that is "almost right, but not quite". Catching those near-misses before users do — with test sets, automated checks and regression tracking — is precisely what an AI engineer is paid for. So if you enjoy building the harness more than writing the prompt, AI engineering will suit you. ## What do official salary figures show? Official statistics cover these roles unevenly, and that matters more than any single number. - **Data scientists (US).** As of September 2026, the latest BLS figures put the median salary at $120,230 (May 2025), with employment projected to grow 35% from 2025 to 2035 ([BLS Occupational Outlook Handbook](https://www.bls.gov/ooh/math/data-scientists.htm)). - **ML and AI research (US).** BLS has no ML engineer or AI engineer occupation. The nearest research category, [computer and information research scientists](https://www.bls.gov/ooh/computer-and-information-technology/computer-and-information-research-scientists.htm), had a median of $140,300 in May 2025, and it typically requires a master's degree. - **India.** No official Indian statistic breaks pay down by these three titles. The India figures in the comparison table are indicative medians for three to five years' experience, so check them against live job postings before relying on them. Three caveats matter more than the numbers themselves. **Location dominates.** The same ML engineering role pays materially differently in Bengaluru, Pune and a tier-two city, and remote roles at foreign companies often sit above all of them. **Company type dominates more.** A data scientist at a product company frequently out-earns an ML engineer at a services company by a wide margin. In other words, the title matters less than who you work for. **Compounding matters most.** The three roles start within a few lakhs of each other, but the spread at ten years is enormous, and it tracks depth rather than the initial choice. Someone genuinely excellent at data science will out-earn a mediocre ML engineer. ## How can you test the three careers in three days? Spend one day doing a small version of each job, then pick the one you would happily keep doing on Monday. 1. **Day one, data science.** Take any public dataset with a business question attached and answer the question. Then write a one-page memo with a recommendation and the reason someone might disagree with you. 2. **Day two, ML engineering.** Take a small trained model. Wrap it in an API, containerise it, deploy it somewhere public, and add one monitoring metric. It does not need to be sophisticated. Notice whether the infrastructure part felt like a chore or a puzzle. 3. **Day three, AI engineering.** Build something small on a foundation model API — a question-answering tool over your own notes will do. Next, write five test cases and check whether your changes make the output better or worse. Notice whether the non-determinism frustrated you or interested you. Whichever day you would happily repeat is your answer. This is a far better signal than any salary table, including ours. ## Who should avoid each path? Each role has a type of person it wears down. - **Avoid data science** if you want your work to ship as software, or if persuading sceptical stakeholders drains you. - **Avoid ML engineering** if you dislike on-call duty, infrastructure and debugging systems you did not write. - **Avoid AI engineering** if constant tooling churn frustrates you, or if you need deterministic, provably correct outputs. ## Where to go from here Once you have picked, the sequencing matters. Our [AI Engineer roadmap](https://www.1stepgrow.com/articles/ai-engineer-roadmap-2026) and [Data Scientist roadmap](https://www.1stepgrow.com/articles/data-scientist-roadmap-2026) both lay out stage-gated paths with portfolio checkpoints, and the [MLOps engineer career guide](https://www.1stepgrow.com/articles/mlops-engineer-career-guide) covers the production side of ML engineering. If you are still weighing whether formal training is worth the cost, read [Is a data science bootcamp worth it](https://www.1stepgrow.com/articles/is-a-data-science-bootcamp-worth-it) first. ## Frequently asked questions ### Can I move from data science to ML engineering later? Yes, and it is the most common transition of the three. The gap is software engineering practice — testing, CI, containers, cloud infrastructure. Most data scientists who make the move spend six to nine months deliberately closing that gap while still employed. ### Is AI engineering just prompt engineering with a better title? No. Prompting is maybe ten percent of it. The job is retrieval architecture, evaluation harnesses that catch regressions, cost and latency management, and guardrails — the same discipline as any other production engineering, applied to a non-deterministic component. ### Which role has the best long-term prospects? ML engineering, on current evidence. It requires the combination of skills that is hardest to automate and hardest to hire for. But 'best prospects' matters less than sustained interest — the person who enjoys the work will out-earn the person who picked the higher median and burned out. ### Do I need a master's degree for any of these? For data science in research-heavy organisations and for AI research roles, often yes. For ML engineering and AI engineering, no — a portfolio of deployed work outweighs a degree in almost every hiring process we examined. ### Which is easiest to break into from a non-technical background? AI engineering, by some distance. The tooling has lowered the floor dramatically and you can ship a genuinely useful product in weeks. The trade-off is that the same low floor means more competition at entry level. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/data-science-vs-machine-learning-vs-ai/). Cite with the title, "1stepGrow" and a link._ --- # Data Analytics Course with Python: Online Options Worth Paying For > How to choose a data analytics course with Python online: five options compared, the small Python subset analysts need, and why SQL still comes first. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Aug 19, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Listicle · **Read time:** 7 min - **Canonical URL:** https://www.1stepgrow.com/data-analytics-course-with-python-online/ ## Key takeaways - Python has moved from differentiator to baseline in analytics postings — but SQL is still what gets tested first. - Learn Python inside an analytics context, not as general programming. The relevant subset is small. - pandas plus one visualisation library covers the overwhelming majority of analytics work. - The free path is genuinely competitive here; paid programs buy structure and support, not secret content. A worthwhile data analytics course with Python teaches the language inside analysis work: pandas, one plotting library and just enough core Python, applied to real datasets. Free courses cover the content well, so pay only when you need structure, feedback or a credential, and practise SQL alongside whichever option you pick. This guide is for aspiring and working analysts who want Python without wasting months on it. The subset an analyst needs is far smaller than a general programming course teaches: you can be productive without writing a class or learning what a decorator is. The common trap is spending eight weeks on general exercises, then meeting a real dataset with no idea where to start. Below, five options are compared on Python depth and price, followed by the exact subset to learn, a syllabus checklist and a list of people who can skip paying altogether. ### 1. Learnbay — Data Analytics Certification Program Python taught inside analytics, with foundation classes for people who have never coded. - **Provider:** Learnbay - **Price:** On request (EMI available) - **Duration:** 9 months - **Level:** Beginner to intermediate - **1stepGrow score:** 4.5/5 - **Best for:** Working professionals who need structure and a credential - **Website:** https://www.learnbay.co/datascience/data-analytics-certification-course **Pros** - Foundation programming classes included rather than assumed - 275 hours across live lectures, practice and interview preparation - IBM-certified capstone projects on real datasets - Repeat-batch access for 2-3 years covers the weeks work eats **Cons** - Nine months is longer than a pure Python course needs to be - Pricing requires an enquiry **Verdict:** The strongest option if you want the whole analytics arc with accountability attached. If you only need Python itself, the cheaper options below will do. ### 2. Python for Data Science, AI & Development A focused, cheap introduction to exactly the Python subset analytics uses. - **Provider:** Coursera (IBM) - **Price:** $49/month (first module free to preview) - **Duration:** 4-6 weeks - **Level:** Beginner - **1stepGrow score:** 4.3/5 - **Best for:** Getting productive in pandas quickly - **Website:** https://www.coursera.org/learn/python-for-applied-data-science-ai **Pros** - Tight scope — no wasted weeks on general programming - First module free to preview; financial aid available - Hands-on from the start **Cons** - Shallow on statistics - No support structure **Verdict:** The most efficient way to go from zero Python to useful pandas. Pair it with SQL practice. ### 3. Python for Everybody The gentlest genuinely good programming introduction there is. - **Provider:** Coursera (University of Michigan) - **Price:** $49/month (first module free to preview) - **Duration:** 6-8 weeks - **Level:** Absolute beginner - **1stepGrow score:** 4.7/5 - **Best for:** People who have genuinely never programmed - **Website:** https://www.coursera.org/specializations/python **Pros** - Assumes nothing at all - Excellent pacing - First module free to preview; financial aid available **Cons** - General programming, not analytics-specific - You will need a second course for pandas **Verdict:** Best starting point if the word 'variable' is unfamiliar. Otherwise start higher. ### 4. Google Data Analytics Professional Certificate Broad analytics grounding, but its programming component is R rather than Python. - **Provider:** Coursera - **Price:** $49/month - **Duration:** 3-6 months - **Level:** Beginner - **1stepGrow score:** 4.4/5 - **Best for:** A recognised first credential - **Website:** https://www.coursera.org/professional-certificates/google-data-analytics **Pros** - Well-structured, employer-recognised - Covers the analytics workflow end to end **Cons** - Teaches R, not Python - Very widely held **Verdict:** Good credential, no Python. Treat Python as something you learn elsewhere. ### 5. pandas documentation + a real dataset Unfashionable advice that works better than most courses. - **Provider:** Free - **Price:** Free - **Duration:** Ongoing - **Level:** All levels - **1stepGrow score:** 4.2/5 - **Best for:** Anyone who learns by doing - **Website:** https://pandas.pydata.org/docs/user_guide/index.html **Pros** - The user guide is genuinely excellent - Working on a dataset you care about sustains motivation **Cons** - No structure, no credential, no feedback **Verdict:** Not a substitute for a course if you need accountability. A large accelerator if you do not. ## Which Python analytics option suits your starting point? The table below summarises the list, using prices and durations as of September 2026. Subscription prices vary by country, so check the course page in your own region. | Option | Python depth | Price | Duration | Start here if… | |---|---|---|---|---| | Learnbay Data Analytics Certification Program | Full analytics arc | On request | 9 months | You want structure and a credential | | IBM Python for Data Science, AI & Development | Focused analytics subset | $49/month (first module free to preview) | 4-6 weeks | You want pandas quickly | | Python for Everybody (University of Michigan) | General programming | $49/month (first module free to preview) | 6-8 weeks | You have never coded | | Google Data Analytics Professional Certificate | None (teaches R) | $49/month | 3-6 months | You want a recognised first credential | | pandas documentation + a real dataset | As deep as you go | Free | Ongoing | You learn by doing | Coursera [replaced free auditing with a free first-module preview](https://blog.coursera.org/introducing-courseras-new-course-preview-experience/) in August 2025, so finishing a Coursera course now needs a paid plan or approved financial aid. One more correction to a common assumption: the Google certificate is not a Python course. Its [Coursera FAQ](https://www.coursera.org/professional-certificates/google-data-analytics) states that the programme teaches R. That still makes it a sound analytics credential, but you will need a separate course for Python. ## What Python does a data analyst actually need? Here is the honest scope of what analytics work actually requires. **Core language.** Variables, lists, dictionaries, loops, conditionals, functions. Roughly two weeks. You will use dictionaries and list comprehensions constantly and almost nothing else from this tier. **pandas.** Loading data, selecting and filtering, `groupby`, merging, handling missing values, reshaping. This is the bulk of the job and deserves the bulk of your time. The official [pandas getting-started tutorials](https://pandas.pydata.org/docs/getting_started/intro_tutorials/index.html) follow almost exactly this sequence, from reading tabular data to combining tables and handling time series. **One visualisation library.** matplotlib for control, seaborn for speed, plotly for interactivity. Learn one properly rather than three superficially — our [plotly](https://www.1stepgrow.com/articles/plotly-tutorial) and [matplotlib](https://www.1stepgrow.com/articles/matplotlib-styling-customization) tutorials cover two of them. **Enough NumPy to not be confused.** Arrays, broadcasting, vectorised operations. You will rarely write raw NumPy, but pandas sits on it and error messages leak through. Our [NumPy series](https://www.1stepgrow.com/articles/numpy-tutorial) covers what you need. That is it. Everything beyond this is nice to have, and none of it is what fails candidates in interviews. **A working setup.** Before week one, get Jupyter running locally so that you can open your own files rather than only the course's hosted notebooks. Our guides to installing Anaconda and Jupyter on [Windows](https://www.1stepgrow.com/articles/install-anaconda-windows) and [macOS](https://www.1stepgrow.com/articles/install-anaconda-mac) take about half an hour. ## How do you judge a data analytics course with Python before enrolling? Course pages all sound alike, so check the syllabus against these points instead: - **When pandas starts.** If it arrives after several weeks of general programming, the course was not written for analysts. - **What the datasets look like.** Clean, tiny tutorial files teach syntax; messy files with missing values teach the job. - **Whether projects end in a finding.** A notebook that stops at a chart is an exercise. By contrast, a project that ends with a written conclusion is a portfolio piece. - **Whether SQL is included.** If it is not, budget time for it separately. - **What feedback you get.** Paid courses justify their price through code review and deadlines, so ask exactly who reviews your work. ## What do analytics employers list as required skills? The official job frameworks back up the "Python plus SQL" advice. For example, the US O*NET profile for [Business Intelligence Analysts](https://www.onetonline.org/link/summary/15-2051.01) names Python, R and SQL among its technology skills, next to Power BI, Tableau and Excel. The core tasks, however, are reports, dashboards and trend analysis. In other words, Python is a tool for that work rather than the work itself. ## Who should skip a paid Python course? Paying is unnecessary for many learners. Skip it if: 1. **You already program in another language.** The free official documentation and a real dataset will get you productive within weeks. 2. **You reliably finish self-paced material.** The free documentation and Coursera's financial aid route cover much of the same content. 3. **Your target role is SQL-heavy reporting.** Then SQL practice deserves the budget first. ## Why should SQL practice run alongside a Python course? Because the interview round that filters analytics candidates is usually a live SQL test, and Python courses rarely prepare you for it. Every Python course will make you feel productive. Almost none will prepare you for the round where analytics hiring actually filters. That round is SQL, it usually comes first, and it is live. Window functions, CTEs, a join with an awkward condition — written while someone watches. The [PostgreSQL tutorial on window functions](https://www.postgresql.org/docs/current/tutorial-window.html) is a good free place to start on the topic candidates find hardest. Run SQL practice in parallel from week one, whatever Python course you take. It is the single highest-return hour in an analytics job search, and the one most consistently under-prepared. ## How do you actually finish a Python course? Work the course material against a messy dataset you care about rather than the tutorial files. The difference between people who finish these courses and people who abandon them is rarely aptitude. Instead, it is whether the practice data is interesting. Public datasets in tutorials are clean, small and boring by design. Find something you actually care about — your running data, your city's transport statistics, your team's ticket exports, anything with real mess in it — and work the course material against that instead. Messy data is where the learning is. It is also where the portfolio piece comes from, because "I analysed this thing I care about and here is what surprised me" is a much better interview story than "I completed the module". ## Related reading For the predictive layer above this, see [advanced data analytics with predictive analytics](https://www.1stepgrow.com/articles/advanced-data-analytics-course). For the career decision, read [Data Analyst vs Data Scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist). ## Frequently asked questions ### Do I need Python for a data analytics job? Increasingly yes. It has moved from a bonus line to an assumed baseline in most postings over the last few years. That said, SQL is still tested first and more heavily — get both, in that order of priority. ### Python or R for analytics? Python, unless you are targeting academia, biostatistics or a specific team that has standardised on R. Python has more job listings and better integration with engineering workflows. ### How much Python do I actually need? Less than a general programming course teaches. Data structures, functions, control flow, and then pandas and one plotting library. You can be genuinely productive without ever writing a class. ### Should I learn pandas or SQL first? SQL. It is tested more often, it is what you will use to get the data before pandas ever touches it, and it is where interviews eliminate candidates. ### Are free Python courses good enough? For the content, yes — the free material in this space is excellent. Paid programs buy structure, feedback and a credential. If you reliably finish things on your own, the free path is genuinely competitive. --- _Source: 1stepGrow (https://www.1stepgrow.com/data-analytics-course-with-python-online/). Cite with the title, "1stepGrow" and a link._ --- # Advanced Digital Marketing Course: Why Measurement Beats Channels > An advanced digital marketing course should teach measurement: attribution, incrementality and SQL. Five options compared, plus a test for outdated syllabi. - **Author:** Nishant Kiran — B2B SaaS & EdTech Content Strategist (https://www.1stepgrow.com/authors/nishant-kiran/) - **Published:** Aug 18, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Digital Marketing · **Format:** Listicle · **Read time:** 8 min - **Canonical URL:** https://www.1stepgrow.com/articles/advanced-digital-marketing-course/ ## Key takeaways - The scarce skill in marketing is measurement, not channel execution. Price your course choice accordingly. - Attribution and incrementality are what separate a senior marketer from a channel operator — and are missing from most syllabi. - AI has commoditised production. It has not commoditised knowing whether the work is paying for itself. - A marketer who can write SQL is in a very small and well-paid group. An advanced digital marketing course should move you from running channels to measuring them: attribution, incrementality testing, experiment design and enough statistics and SQL to defend your numbers. That is why the shortlist below mixes marketing programmes with analytics training, because measurement is the skill that is now scarce. Marketing work has split into execution and measurement. Generative tools have compressed the value of writing copy and shipping creative, while deciding whether any of it worked has become harder to hire for, because there is far more output to evaluate. Many "advanced" courses still teach the first half, and some are simply out of date: Google retired four GA4 attribution models in November 2023, so any syllabus that still teaches them as current has not been updated. This shortlist is for marketers with a few years of channel experience who want to own the numbers. It compares five options, lists the four topics that prove a course is serious, and explains why SQL belongs on a marketing reading list. ### 1. Advanced Google Analytics & Attribution The measurement layer, taught by the people who built the measurement layer. - **Provider:** Google Skillshop + independent courses - **Price:** Free - **Duration:** 4-8 weeks - **Level:** Intermediate - **1stepGrow score:** 4.3/5 - **Best for:** Marketers who need to defend their numbers - **Website:** https://skillshop.withgoogle.com/ **Pros** - Free and directly applicable - GA4 coverage is authoritative - Certification recognised across agencies **Cons** - Google-specific by definition - Thin on statistical reasoning behind attribution **Verdict:** Start here. It is free, it is the industry baseline, and its absence is noticed in interviews. ### 2. Learnbay — Data Analytics Certification Program The analytics depth marketing courses skip — SQL, statistics and modelling, taught for non-programmers. - **Provider:** Learnbay - **Price:** On request (EMI available) - **Duration:** 9 months - **Level:** Beginner to intermediate - **1stepGrow score:** 4.5/5 - **Best for:** Marketers who want to become the person who owns measurement - **Website:** https://www.learnbay.co/datascience/data-analytics-certification-course **Pros** - Teaches SQL and statistics properly, which marketing programs almost never do - Built for non-programmers, so no coding background needed - IBM-certified capstone projects you can run on your own campaign data - Weekend batches designed around a working week **Cons** - Not a marketing course — you bring the marketing context yourself - Nine months is a real commitment **Verdict:** An unusual recommendation on a marketing list, and a deliberate one. The bottleneck for senior marketers is analytical capability, not another channel tactic — and a marketer who can query the warehouse directly is rare and valuable. ### 3. CXL Digital Analytics & Growth Minidegrees The most rigorous marketing training available, priced for teams. - **Provider:** CXL - **Price:** $300-$1,000+/year - **Duration:** 3-6 months - **Level:** Advanced - **1stepGrow score:** 4.5/5 - **Best for:** Experienced marketers going deep on experimentation - **Website:** https://cxl.com/institute/ **Pros** - Genuinely advanced — experimentation, statistics, research methods - Taught by recognised practitioners - Strong on conversion optimisation **Cons** - Expensive for an individual - Assumes real marketing experience **Verdict:** The best pure-marketing option on this list if your employer will fund it. ### 4. Meta & Google Ads Certifications Platform credentials — necessary, and not sufficient. - **Provider:** Meta Blueprint / Google Skillshop - **Price:** Free-$150 - **Duration:** 2-6 weeks - **Level:** Beginner to intermediate - **1stepGrow score:** 3.8/5 - **Best for:** Agency roles and clearing screening filters **Pros** - Free or cheap, quick to complete - Frequently required by agencies **Cons** - Tests platform knowledge, not marketing judgement - Expires and needs renewal **Verdict:** Get them because their absence is a filter. Do not expect them to differentiate you. ### 5. SQL for Marketers The single most under-rated skill in marketing. - **Provider:** Various (Mode, DataCamp, independent) - **Price:** Free-$300 - **Duration:** 4-8 weeks - **Level:** Beginner - **1stepGrow score:** 4.4/5 - **Best for:** Marketers who are tired of waiting on the data team **Pros** - Removes your dependency on analyst availability - Immediately useful - Small time investment for a large capability jump **Cons** - Not a marketing qualification **Verdict:** Eight weeks that will change how you work. Very few marketers do this, which is exactly why it pays. ## How does each option on the shortlist compare? Here is the list in one view, using the prices and durations above as of September 2026. Because platform fees and certification rules change, check the provider's page before you commit. | Option | What it teaches | Price | Duration | Level | Best for | |---|---|---|---|---|---| | Advanced Google Analytics & Attribution (Skillshop) | GA4 and attribution setup | Free | 4-8 weeks | Intermediate | Defending your numbers | | Learnbay Data Analytics Certification Program | SQL, statistics, modelling | On request | 9 months | Beginner to intermediate | Owning measurement for a team | | CXL Digital Analytics & Growth Minidegrees | Experimentation, research, CRO | $300-$1,000+/year | 3-6 months | Advanced | Experienced marketers | | Meta & Google Ads certifications | Platform knowledge | Free-$150 | 2-6 weeks | Beginner to intermediate | Agency screening filters | | SQL for marketers | Querying your own data | Free-$300 | 4-8 weeks | Beginner | Escaping the data team's queue | Notice that only two rows are marketing courses in the usual sense. The rest teach the analytical layer, since that is where most marketing syllabi stop short. ## What should an advanced digital marketing course teach? Scan any advanced digital marketing syllabus for these four topics. Their presence tells you whether the course is serious. **Attribution, honestly.** Not "here is how to configure a model in GA4" but why last-click is wrong, what multi-touch actually assumes, and why the numbers still will not reconcile with the platform's own reporting. Also check that the material is current. Google's help page on [GA4 attribution](https://support.google.com/analytics/answer/10596866) confirms that the first click, linear, time decay and position-based models stopped being available in November 2023, so a course still teaching them as GA4 options has not been updated. **Incrementality.** The question that matters: would this conversion have happened anyway? Geo holdouts, PSA tests, matched-market design. Almost no marketing course teaches this, and it is the difference between reporting spend and justifying it. For example, Google's own [Conversion Lift](https://support.google.com/google-ads/answer/12003020) tool splits an audience into a group that sees ads and a control group that does not, then measures the difference. **Experiment design.** Sample size, test duration, and why calling a test after three days because it looks good is how marketing teams convince themselves of things that are not true. **Statistics for marketers.** Enough to know when a difference is noise. This is the foundation the other three sit on. A course covering all four is worth real money. A course covering channel tactics and calling it advanced is not. ## What do measurement-focused marketing roles involve? They centre on analysing data and forecasting, and official job descriptions confirm the shift. The US Bureau of Labor Statistics' profile of [market research analysts](https://www.bls.gov/ooh/business-and-financial/market-research-analysts.htm) lists monitoring and forecasting marketing and sales trends, analysing data with statistical software, and turning findings into tables, graphs and reports. It puts median US pay at $78,760 in May 2025. Those duties map directly onto the four topics above. In other words, a course that never makes you analyse data is preparing you for a narrower role than the one you probably want. ## Why should marketers learn SQL? Because it removes the wait for the data team, and that wait is the real bottleneck. SQL looks out of place on a marketing page, yet it is the most consistent advice we give marketers. The bottleneck in most marketing teams is not ideas: every question requiring data goes into the analytics team's queue and comes back in four days, by which point the decision has been made without it. A marketer who can write their own queries closes that loop themselves. They ask better questions because the cost of asking dropped to nothing, they catch bad numbers before presenting them, and they end up owning measurement for the team — which is the path to senior roles. It takes about eight weeks. Very few marketers do it. Those two facts together are the entire argument. ## Who should not buy an advanced digital marketing course yet? Advanced material is wasted without the base underneath it. Hold off if any of these apply: - **You have never run a paid campaign with your own budget.** Platform certifications and a small live test will teach you more first. - **You cannot yet read a GA4 report confidently.** Start with the free Skillshop training, then come back. - **You want to specialise in one channel.** In that case, a channel course fits better; our [digital marketing specialisation guide](https://www.1stepgrow.com/digital-marketing-specialization-course) compares SEO, SMM and SEM routes. - **Your employer will not give you data access.** Then the analytics skills have nowhere to be practised, so fix that conversation first. ## How do you build a marketing portfolio that gets interviews? Show two or three case studies with numbers, including one that failed. Marketing hiring is unusual in that a portfolio is both expected and rarely done well. The weak version is a list of campaigns and impressions. The strong version is two or three case studies structured as: here is what we needed to move and why, here is what I did, here is what happened, here is what I would change. Include one that did not work. A marketer who can analyse their own failed campaign honestly is signalling exactly the judgement senior roles require. Yet almost nobody does it, which makes it disproportionately memorable. ## Where to go next Open the syllabus of any course you are considering and search it for "incrementality" and "experiment"; if neither appears, keep looking. For the specialist channel route, see our [digital marketing specialisation guide](https://www.1stepgrow.com/digital-marketing-specialization-course). For the intensive option, [the digital marketing bootcamp guide](https://www.1stepgrow.com/digital-marketing-course-online) covers brand-audit-led programs. If the SQL argument convinced you, the [advanced data analytics course guide](https://www.1stepgrow.com/articles/advanced-data-analytics-course) shows where that path leads. ## Frequently asked questions ### Is digital marketing still a good career with AI generating content? Content production has genuinely been commoditised. Strategy, measurement and knowing whether the spend is paying for itself have not — and are in more demand, because there is now far more output to evaluate. Move toward the measurement end and the picture looks good. ### Do digital marketers need to learn SQL? Not to do the job, but it is one of the highest-leverage skills available to a marketer. It removes your dependence on the analytics team's queue and lets you answer your own questions in minutes rather than days. It also lets you catch a wrong number before you present it, and a focused course takes around four to eight weeks. ### What is the difference between digital marketing and marketing analytics? Digital marketing executes across channels; marketing analytics decides whether the execution worked and where the next rupee should go. The analytics side leans on attribution, incrementality testing, experiment design and statistics rather than channel tactics. Those skills are harder to find, which is why measurement-focused marketers tend to be harder to hire. ### Are marketing certifications worth it? Platform certifications from Google and Meta are cheap, quick and frequently required by agencies, so get them. They will not differentiate you, because they test platform knowledge rather than judgement. A documented campaign with real numbers and an honest post-mortem will, and so will evidence that you can measure incrementality rather than just report spend. ### What should a portfolio look like for marketing roles? Two or three campaigns with the numbers attached: what you set out to move, what you did, what happened, and what you would change. Including one that underperformed, analysed honestly, is more persuasive than three clean wins. Keep each case study to one page, and state how you knew the result was caused by your work. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/advanced-digital-marketing-course/). Cite with the title, "1stepGrow" and a link._ --- # Digital Marketing Specialisation Courses: SEO, SMM and SEM Compared > Choosing a digital marketing specialisation? SEO, SEM and social media compared on feedback speed, measurement, AI exposure and a cheap way to test each. - **Author:** Nishant Kiran — B2B SaaS & EdTech Content Strategist (https://www.1stepgrow.com/authors/nishant-kiran/) - **Published:** Aug 17, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Digital Marketing · **Format:** Comparison · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/digital-marketing-specialization-course/ ## Key takeaways - SEM has the clearest measurement and the fastest feedback loop, which makes it the easiest specialisation to prove value in. - SEO has the highest ceiling and the longest proof cycle — and has been the most disrupted by AI search. - Social media marketing has the lowest barrier to entry and, consequently, the most competition at junior level. - Whichever you pick, the analytics layer underneath is what separates a practitioner from an operator. The right digital marketing specialisation depends on the feedback loop you can live with. SEM (paid search) gives the fastest, clearest proof of your impact; SEO has the highest long-term ceiling but takes months to show results; social media marketing is the easiest to enter and the most crowded at junior level. Generalist digital marketing training has become a commodity, so specialising is how most marketers move up. The catch is that SEO, SEM and social are different jobs with different economics. Pick on instinct and you can spend two years waiting months for SEO results you lack the patience for, or fighting for crowded junior social roles. This comparison is for new and early-career marketers choosing where to go deep. It sets out what each job is like week to week, which certifications count, how AI search and automation have hit each one, and a month-long way to test all three before paying for a course. | Feature | SEO | SEM / Paid Search | Social Media Marketing | | --- | --- | --- | --- | | Time to demonstrate results | 4-9 months | Days to weeks | Weeks to months | | Measurement clarity | Muddy — attribution is hard | Excellent — spend to conversion | Poor to moderate | | Barrier to entry | Moderate | Moderate — real budget needed to practise | Low | | Competition at junior level | Moderate | Lower | Very high | | Exposure to AI disruption | High — AI search changes the surface | Moderate — automation of bidding | High — content is commoditised | | Ceiling for senior roles | High — owns a durable asset | High — owns real budget | Moderate | | Technical depth required | High for technical SEO | Moderate, heavy on analytics | Low to moderate | | Practise without an employer | Yes — build your own site | Hard — needs ad budget | Yes — build your own audience | | Transfers to other roles | Content strategy, product | Analytics, growth | Brand, community | **Verdict — SEM / Paid Search:** For someone choosing a specialisation today, SEM offers the best combination of fast feedback, clear measurement and defensible value — you can prove what you contributed in a way SEO and social rarely allow. That said, SEO has the higher long-term ceiling because organic visibility is an asset the company owns rather than rents, and the people who can navigate its AI-driven disruption will be scarce and well paid. A note on the pay ranges above: no official Indian source publishes salary bands by marketing specialisation, so treat them as indicative ranges rather than survey data, and check current job postings in your city. ## What is each specialisation like day to day? SEO is slow and systematic, SEM is fast and numerical, and social is creative but hard to measure. **SEO** is a long game against a system you do not control and cannot fully observe. The work is content strategy, technical health, internal linking and earning authority. The satisfying part is that it compounds: traffic you earned last year is still arriving. The frustrating part is a feedback loop measured in months, which makes it hard to know whether you are good at it early on. **SEM** is the most immediately legible marketing job there is: spend money, read the data, adjust. Within weeks you know whether your decisions worked. That clarity is why it is the easiest specialisation to build a case for a raise in, and why analytical people tend to thrive in it. **Social media marketing** is creative work with a distribution problem attached. The platforms change the rules regularly and without warning, organic reach for brands keeps declining, and measurement is genuinely difficult. It has the lowest barrier to entry, which is exactly why junior competition is fierce. ## Digital marketing specialisation skills compared Each specialisation rests on a different daily toolkit. This table shows what you would actually spend your week doing, and where the free official training lives. | | SEO | SEM / paid search | Social media marketing | |---|---|---|---| | Daily work | Content briefs, technical fixes, link earning | Bids, budgets, ad copy tests, conversion tracking | Content calendars, community replies, paid social | | Core skill | Understanding how search engines read a site | Unit economics of paid traffic | Knowing what a specific audience responds to | | Free official starting point | [Google's SEO Starter Guide](https://developers.google.com/search/docs/fundamentals/seo-starter-guide) | [Google Ads certifications](https://support.google.com/google-ads/answer/9702955?hl=en) on Skillshop | Platform help centres and [HubSpot Academy](https://academy.hubspot.com/courses) | | Recognised credential | None from Google | Google Ads Search, Display, Video, Shopping and Apps | No single industry standard | | How to practise alone | Build and rank your own site | Run a small campaign with your own budget | Grow an account around one topic | ## Which certifications matter for each specialisation? Certifications matter most in SEM, where the platforms issue them, and least in SEO, where they do not. - **SEM.** Google offers five Google Ads certifications: Search, Display, Video, Shopping ads and Apps. As of September 2026 each needs a score of at least 80% on a timed assessment, lasts one year, and counts toward an agency's Google Partner status. Agencies therefore ask for them routinely. - **SEO.** Google publishes no SEO certification. Instead, its starter guide states that there are no secrets that will automatically rank a site first, and a site you ranked yourself remains the strongest credential. - **Social media marketing.** Platform and vendor courses exist, but no credential carries the weight of a Google Ads certification. A portfolio of accounts you grew, with numbers, does more. ## How has AI changed SEO, SEM and social media marketing? All three have been affected, and not equally: SEO most at the surface, SEM least, and social through commoditised content production. **SEO** has been hit hardest at the surface level. AI-generated answers absorb informational queries that used to produce clicks. But the effect is uneven: commercial-intent searches still convert through traditional results, brand searches are unaffected, and being the source an AI system cites is becoming its own discipline. Google says there are [no additional requirements or special optimisations](https://developers.google.com/search/docs/appearance/ai-features) needed to appear in AI Overviews or AI Mode, so the fundamentals still apply. Google also reports that clicks from results pages with AI Overviews are higher quality, meaning visitors spend more time on the site; that is Google's own claim, so test it against your analytics. **SEM** is the most insulated. Bidding automation has been eating the manual work for years, which has pushed the job toward strategy, measurement and creative testing, a more interesting job than the one it replaced. **Social** has seen production commoditised faster than anything else. Anyone can generate a month of posts now. What has not been commoditised is knowing what to say, to whom, and whether it moved anything. The pattern is consistent across all three: execution is being automated, judgement is not. ## Which skill matters whichever specialisation you choose? Analytics. Whichever specialisation you choose, the measurement layer is what separates a practitioner from an operator. An operator runs the campaigns, while a practitioner can tell you whether the campaigns worked, distinguish a real lift from noise, and defend the answer to a sceptical finance team. That capability, meaning attribution, experiment design and enough statistics to know when a difference is meaningful, is portable across all three specialisations. It also keeps compounding as the tactical layer keeps changing underneath you. Our [advanced digital marketing and analytics guide](https://www.1stepgrow.com/articles/advanced-digital-marketing-course) covers the programs that teach it. ## How can you test a specialisation before committing? Run a small, cheap experiment in each one over about a month before paying for a course: 1. **SEO, in one month.** Publish ten useful pages on a small site, submit it to Google Search Console, and watch which pages earn impressions. 2. **SEM, in two weeks.** Pass the Google Ads Search certification, then run a tiny campaign with a budget you can afford to lose and track cost per conversion. 3. **Social, in one month.** Post consistently on one platform about one topic and record which formats earn saves and shares, not just likes. 4. **Compare how each felt.** The specialisation whose waiting and whose data you enjoyed is usually the right one. ## Common mistakes when picking a specialisation - **Choosing by starting salary alone.** The ceiling and the feedback loop matter more over five years. - **Collecting certificates in all three.** Early on, depth in one gets you hired far more reliably. - **Ignoring measurement.** Every specialisation now rewards people who can prove impact. - **Assuming AI has ended SEO.** It has changed what winning looks like, not removed the work. - **Practising SEM with no budget.** Without real spend you cannot show real results, so plan a small test budget. ## How to choose this week Pick the one whose feedback loop you can tolerate. If waiting six months to find out whether you were right sounds intolerable, do not choose SEO, because you will burn out before the results arrive. Arguing from a spreadsheet may sound like the good part of the job rather than the tedious part; if so, SEM will suit you. Already making things people want to look at is a real advantage in social. Pair it with measurement skills, though, or you will hit the ceiling fast. ## Where to go next Start the two-week SEM test or the one-month SEO test this week, and note how the waiting feels. For the bootcamp route, see our [digital marketing bootcamp guide](https://www.1stepgrow.com/digital-marketing-course-online). For the analytics depth underneath all three, read [advanced digital marketing and analytics courses](https://www.1stepgrow.com/articles/advanced-digital-marketing-course). For more head-to-heads, browse the [comparisons hub](https://www.1stepgrow.com/comparisons). ## Frequently asked questions ### Which digital marketing specialisation pays the most? At senior levels SEO and SEM are comparable and both above social media marketing, because both come to control significant budget or significant traffic. At junior level SEM often edges ahead, largely because fewer people can demonstrate competence with real spend. ### Has AI search killed SEO? It has changed what winning looks like rather than removing the work. Zero-click answers reduce informational traffic, which pushes value toward commercial-intent queries, brand strength and being the source AI systems cite. That is a harder job than before and a more valuable one. ### Can I specialise in more than one? Eventually, and most senior marketers do. Early on, depth in one gets you hired far more reliably than familiarity with three — the interview will test what you claim. Pick one specialisation for your first two years, learn the analytics underneath it, and broaden once you can show results. ### Do I need a certification to get an SEO or SEM job? For SEM, get them: Google Ads certifications are free on Skillshop, last one year and count toward an agency's Google Partner status. For SEO there is no certification from Google and no meaningful industry standard; a site you ranked yourself is the credential, and it is a much better one than any certificate. ### Which is easiest to break into without experience? Social media marketing has the lowest barrier — and consequently the most competition and the lowest starting pay. SEO is the easiest to practise independently, because you can build and rank your own site without anyone's budget. SEM is the hardest to practise alone, because you need real ad spend to show real results. --- _Source: 1stepGrow (https://www.1stepgrow.com/digital-marketing-specialization-course/). Cite with the title, "1stepGrow" and a link._ --- # Best Cloud Certification in 2026: 11 Ranked by Cost, Prep and Renewal > The best cloud certification in 2026 depends on your employer's cloud. Eleven AWS, Azure, GCP and Kubernetes options ranked by fee, prep time and renewal. - **Author:** Sivaranjani S — Cloud Operations Engineer, Zoho (https://www.1stepgrow.com/authors/sivaranjani-s/) - **Published:** Aug 16, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Cloud Computing · **Format:** Listicle · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/top-cloud-certifications-2026/ ## Key takeaways - Associate-level certifications clear HR filters. Professional-level ones actually change what you are paid. - The certification only matters in a shop that uses that cloud — pick by employer, not by market share. - Total cost is rarely the exam fee. Budget for labs, practice exams and renewal, which comes round every one to three years depending on the vendor. - For anyone entering the field, one associate cert plus one deployed project beats three certs and no portfolio. The best cloud certification for you is the one matching the cloud your target employers already run. For most newcomers that means an associate exam such as AWS Solutions Architect – Associate or Azure Administrator; for experienced engineers, a professional or expert credential moves pay. Below we rank eleven options by cost, study time and renewal burden. Hiring managers will tell you they do not care about cloud certifications, and then filter their candidate list by them anyway. Both things are true: a certification rarely wins you a job, but its absence quietly loses you the interview you never heard about. The expensive mistake is the opposite one, collecting badges for a cloud your target employers do not use. This ranking is for developers, sysadmins and data engineers choosing a first or next credential. It lists official fees, exam length and validity for every exam, what renewal costs on each vendor, and a starting-point table that tells you which exam to book first. We ranked the eleven that matter by four things: what it costs in total, how long it genuinely takes to prepare for, how much salary movement it correlates with, and what twenty-three hiring managers told us they actually screen for. ### 1. Learnbay - AI-Powered Cloud & DevOps Master Program The hands-on programme that carries you through the certifications below - and gives you the deployed work they do not test. - **Provider:** Learnbay - **Price:** On request (EMI available) - **Duration:** 10 months - **Level:** Intermediate - **1stepGrow score:** 4.5/5 - **Best for:** Infrastructure and operations engineers moving into cloud platform roles - **Website:** https://www.learnbay.co/cloud-and-devops/ai-powered-cloud-devops-masters-certification **Pros** - 12+ real domain projects across cloud platforms, so you finish with deployed architecture rather than exam recall - CI/CD, infrastructure as code and observability taught together rather than as separate modules - AI-assisted operations woven through, which is where platform work is actually heading - Mock interviews and scheduled job calls, plus IBM certification **Cons** - Ten months against a six-week exam prep is a different kind of commitment - Does not replace the vendor exam itself, which you still sit separately **Verdict:** Our top pick, and it sits at number one for the reason this article keeps returning to: a certification proves recall, a deployed architecture proves capability. This is the only entry that produces the second. Pair it with whichever vendor exam matches your employer and you have both halves. ### 2. AWS Certified Solutions Architect — Professional The single highest-signal cloud credential in the market, and the hardest to fake. - **Provider:** Amazon Web Services - **Price:** $300 exam - **Duration:** 3-5 months prep - **Level:** Advanced - **1stepGrow score:** 4.8/5 - **Best for:** Engineers with 2+ years of hands-on AWS aiming at architect or lead roles **Pros** - Consistently correlates with the largest salary movement of any cloud certification - Scenario-based questions are hard to pass without real experience - Recognised everywhere, regardless of which cloud the employer uses **Cons** - Brutal exam — pass rates are low and retakes are common - Requires genuine production experience; study alone rarely gets you there **Verdict:** If you already work in AWS daily, this is the best return on three months of evenings available anywhere. ### 3. AWS Certified Solutions Architect — Associate The default first cloud certification, and still the right one for most people. - **Provider:** Amazon Web Services - **Price:** $150 exam - **Duration:** 6-10 weeks prep - **Level:** Intermediate - **1stepGrow score:** 4.6/5 - **Best for:** Anyone entering cloud engineering or moving from on-premise infrastructure **Pros** - Enormous volume of high-quality free and cheap study material - Broad coverage that genuinely teaches you the platform - Most frequently listed cert in Indian and global cloud job posts **Cons** - Very widely held — it is a floor, not a differentiator - Question style rewards exam technique alongside knowledge **Verdict:** The correct starting point for the majority of readers. Follow it with a deployed project, not a second cert. ### 4. Google Cloud Professional Cloud Architect The best-designed exam of the three, testing judgement over recall. - **Provider:** Google Cloud - **Price:** $200 exam - **Duration:** 2-4 months prep - **Level:** Advanced - **1stepGrow score:** 4.5/5 - **Best for:** Engineers in GCP shops, and anyone in data-heavy or ML-adjacent infrastructure **Pros** - Case-study format tests real architectural reasoning - Strong signal in data and ML platform teams - Excellent official learning path **Cons** - Smaller job market than AWS outside specific sectors - Two-year validity, so renewal comes round sooner than AWS's three years **Verdict:** The best exam of the three. Whether it is the best certification depends entirely on your target employers. ### 5. Microsoft Certified: Azure Solutions Architect Expert (AZ-305) The enterprise passport. Unglamorous, and quietly very valuable. - **Provider:** Microsoft - **Price:** $165 exam - **Duration:** 2-3 months prep - **Level:** Advanced - **1stepGrow score:** 4.4/5 - **Best for:** Anyone working in or targeting large enterprises, BFSI or government **Pros** - Dominant in enterprise environments, especially in regulated industries - Free, genuinely good official learning materials - Frequently tied to formal internal salary bands **Cons** - Requires the AZ-104 prerequisite, adding cost and time - Heavy service-name recall in the question set **Verdict:** If your employer runs Microsoft, this is the highest-leverage certification on the list for you. ### 6. Certified Kubernetes Administrator (CKA) Not cloud-specific, and more respected by engineers than any vendor badge. - **Provider:** CNCF / Linux Foundation - **Price:** $445 (includes one retake) - **Duration:** 2-3 months prep - **Level:** Intermediate to advanced - **1stepGrow score:** 4.7/5 - **Best for:** Platform, DevOps and SRE engineers **Pros** - Entirely hands-on — you fix real clusters against a clock - Cloud-agnostic, so the value transfers when you change jobs - Genuinely hard to pass without the underlying skill **Cons** - Time pressure in the exam catches out otherwise capable engineers - Two-year validity, renewed by passing the exam again **Verdict:** The most honest certification in this list. Passing it means you can actually do the thing. ### 7. AWS Certified DevOps Engineer — Professional For the people who own the pipeline and carry the pager. - **Provider:** Amazon Web Services - **Price:** $300 exam - **Duration:** 3-4 months prep - **Level:** Advanced - **1stepGrow score:** 4.4/5 - **Best for:** Engineers already running CI/CD and infrastructure as code in production **Pros** - Directly aligned with well-paid platform engineering roles - Covers observability and incident response, not just deployment **Cons** - Assumes substantial prior AWS depth - Narrower audience than the architect track **Verdict:** Excellent second professional certification. A poor first one. ### 8. Microsoft Azure Administrator Associate (AZ-104) The workhorse entry credential for enterprise cloud operations. - **Provider:** Microsoft - **Price:** $165 exam - **Duration:** 6-8 weeks prep - **Level:** Intermediate - **1stepGrow score:** 4.2/5 - **Best for:** Sysadmins and support engineers moving into cloud operations **Pros** - Very achievable for anyone with infrastructure background - Free official learning path of high quality - Often the gate to internal role changes **Cons** - Operations-focused, with limited architectural depth - Little value outside Microsoft environments **Verdict:** The most realistic first step for an on-premise administrator moving to cloud. ### 9. Google Cloud Professional Data Engineer Where cloud certification and data careers overlap most profitably. - **Provider:** Google Cloud - **Price:** $200 exam - **Duration:** 2-3 months prep - **Level:** Advanced - **1stepGrow score:** 4.3/5 - **Best for:** Data engineers on GCP, and analytics engineers moving toward platform work **Pros** - Strong, specific signal for data platform roles - Covers streaming and batch architecture properly **Cons** - Narrow applicability outside GCP data teams **Verdict:** Highly valuable within its niche, close to irrelevant outside it. ### 10. AWS Certified Cloud Practitioner A vocabulary course with an exam attached. - **Provider:** Amazon Web Services - **Price:** $100 exam - **Duration:** 2-3 weeks prep - **Level:** Beginner - **1stepGrow score:** 3.4/5 - **Best for:** Non-engineers — sales, project managers, founders — who need cloud literacy **Pros** - Cheap, fast and genuinely useful for understanding the language - Good confidence builder before the associate exam **Cons** - Carries essentially no weight in engineering hiring - Often sold as a career-changing credential, which it is not **Verdict:** Fine for what it is. Do not let anyone sell it to you as a route into an engineering job. ### 11. HashiCorp Certified: Terraform Associate Small scope, low cost, disproportionate practical value. - **Provider:** HashiCorp - **Price:** $70.50 exam - **Duration:** 3-4 weeks prep - **Level:** Intermediate - **1stepGrow score:** 4.1/5 - **Best for:** Anyone writing infrastructure as code, on any cloud **Pros** - Cheapest certification on this list by a distance - Cloud-agnostic and immediately applicable at work - Short preparation time **Cons** - Narrow — it certifies one tool - Not a substitute for a platform certification **Verdict:** The best value-for-money credential here. Take it alongside something bigger, not instead of it. ## Cloud certification exam facts at a glance The table below lists the official fee, exam length and validity for each vendor exam in the ranking, as of September 2026. Fees are US prices before tax; Microsoft and some other vendors adjust the price by country. | Certification | Exam fee | Exam length | Valid for | |---|---|---|---| | [AWS Solutions Architect – Professional](https://aws.amazon.com/certification/certified-solutions-architect-professional/) | $300 | 180 minutes, 75 questions | 3 years | | [AWS Solutions Architect – Associate](https://aws.amazon.com/certification/certified-solutions-architect-associate/) | $150 | 130 minutes, 65 questions | 3 years | | [Google Cloud Professional Cloud Architect](https://cloud.google.com/learn/certification/cloud-architect) | $200 | 2 hours, 50–60 questions | 2 years | | [Azure Solutions Architect Expert (AZ-305)](https://learn.microsoft.com/en-us/credentials/certifications/azure-solutions-architect/) | $165 | 100–120 minutes | 1 year, free renewal | | [Certified Kubernetes Administrator](https://training.linuxfoundation.org/certification/certified-kubernetes-administrator-cka/) | $445, one retake included | 2 hours, hands-on | 2 years | | AWS DevOps Engineer – Professional | $300 | 180 minutes, 75 questions | 3 years | | Azure Administrator Associate (AZ-104) | $165 | 100 minutes | 1 year, free renewal | | Google Cloud Professional Data Engineer | $200 | 2 hours | 2 years | | AWS Cloud Practitioner | $100 | 90 minutes, 65 questions | 3 years | | [HashiCorp Terraform Associate (004)](https://developer.hashicorp.com/certifications/infrastructure-automation) | $70.50 | 1 hour | 2 years | Microsoft's own FAQ says its associate and expert exams ["typically cost US$165"](https://learn.microsoft.com/en-us/credentials/certifications/frequently-asked-questions) but are priced by country, so check the exam page in your region before you budget. ## How does renewal work for each cloud certification? AWS renews every three years by exam, Microsoft every year for free, and Google Cloud every two years. Renewal is the cost most people forget. Each vendor handles it differently, and the difference adds up over five years. - **AWS.** Certifications last three years. You [renew](https://aws.amazon.com/certification/recertification/) by passing the current exam or a higher-level one; for example, passing Solutions Architect – Professional automatically recertifies the Associate. AWS also offers a one-year extension through a paid Skill Builder subscription. - **Microsoft.** Role-based and expert certifications expire after one year, but [renewal is free](https://learn.microsoft.com/en-us/credentials/certifications/renew-your-microsoft-certification). You pass a short, open-book online assessment during a six-month window before expiry. Fundamentals certifications do not expire. - **Google Cloud.** Professional certifications last two years. You can retake the full exam, sit a one-hour renewal exam ($100 for Cloud Architect), or complete designated courses for a one-year extension. - **CNCF and HashiCorp.** CKA and Terraform Associate both last two years, and renewal means passing the exam again. ## How to choose the best cloud certification for your starting point Start from your current role, then check which cloud your target employers run. This table shows the sensible first move for the most common starting points. | Your starting point | First certification | Next step | |---|---|---| | On-premise sysadmin in a Microsoft shop | Azure Administrator (AZ-104) | AZ-305, which requires AZ-104 first | | Developer or support engineer, cloud open | AWS Solutions Architect – Associate | Terraform Associate and one deployed project | | Two or more years of daily AWS work | AWS Solutions Architect – Professional | DevOps Engineer – Professional if you own pipelines | | Data engineer on Google Cloud | Professional Data Engineer | Professional Cloud Architect | | Platform or SRE engineer on any cloud | Certified Kubernetes Administrator | The architect exam for your employer's cloud | | Non-technical role beside cloud teams | AWS Cloud Practitioner | Stop there unless your role changes | ## What does a cloud certification really cost? Expect roughly $200 to $400 on official fees and training alone, depending on the path. Price rarely decides the best cloud certification, but total cost is rarely the exam fee alone. Here are three realistic paths, using only the official figures above: 1. **AWS associate path.** The $150 exam plus two months of AWS Skill Builder at [$29 a month](https://aws.amazon.com/training/digital/) comes to about $208. AWS also gives certification holders a 50% discount on their next exam, which halves the cost of a second certification. 2. **Azure expert path.** AZ-104 and AZ-305 at $165 each total $330 in the US. Annual renewal then costs nothing but a few hours. 3. **Google Cloud architect path.** The $200 exam plus a $100 renewal exam every two years comes to $400 over four years. Free official training exists for all three. AWS lists more than 1,000 free learning resources on Skill Builder, and Microsoft Learn training is free. ## Should you choose the cloud or the certification first? Choose the cloud first. The single most common mistake is choosing a certification by market share. Market share is irrelevant to you. What matters is which cloud the companies you want to work for actually run. A quick way to check: search your target job titles in your city, open twenty postings, and count the cloud mentions. If eighteen say Azure, the AWS certification is not your best move regardless of what any global ranking says. Regional patterns are strong and stable. Indian services companies and large enterprises skew Microsoft. Product companies and startups skew AWS. Data-heavy and ML-adjacent organisations skew Google Cloud. Regulated industries skew Microsoft almost everywhere. ## Which cloud exam changes in late 2026? AWS is updating the Solutions Architect – Professional exam. Registration for the new SAP-C03 version opens on 27 October 2026, and the last day to sit the current SAP-C02 is 17 November 2026. If you are studying now, either book before that date or switch to the new exam guide. ## Common mistakes with cloud certifications These mistakes came up again and again in our hiring manager interviews: - **Collecting foundational badges.** Cloud Practitioner followed by three associate exams proves recall, not ability. - **Studying only from dumps.** Scenario questions change, and interviewers ask follow-up questions dumps cannot answer. - **Letting a certification lapse.** Microsoft's annual renewal is free, yet a lapsed badge still looks careless on a profile. - **Ignoring hands-on exams.** CKA is performance-based, so reading alone will not get you through it. - **Skipping the project.** Without a deployed architecture, the certification cannot survive the technical interview. ## Why do certification collectors stall? Collectors stall because a stack of badges produces no evidence they can build anything. There is a failure mode we saw repeatedly in the interviews for this piece: the collector. Someone starts with the Cloud Practitioner, passes it, enjoys the feeling, moves to the Associate, passes that, adds the Terraform Associate, then the Azure equivalent. Eighteen months and four certifications later they are still in the same role, because nothing in that sequence produced evidence they can build anything. The alternative that works consistently is one certification, then a project you deploy publicly and write about. A small architecture with infrastructure as code, monitoring and a written explanation of what you would do differently at ten times the traffic. That combination gets past both filters, the automated one and the human one. ## What to do this quarter Working in infrastructure already: go straight for a professional or expert-level certification in the cloud you use. That is where the salary movement is. Entering cloud from development or support: one associate certification, then Terraform Associate if you have a spare month, then build and deploy something real. Not technical, but working alongside cloud teams: Cloud Practitioner is genuinely fine for you. Take it, and then stop. ## Where to go next Open twenty postings for your target role this week and book the matching exam from the starting-point table. For a deeper head-to-head on the two biggest platforms, see [AWS vs Azure certification](https://www.1stepgrow.com/articles/aws-vs-azure-certification-path). If you are building toward a platform engineering role, the [full stack developer roadmap](https://www.1stepgrow.com/articles/full-stack-developer-roadmap-2026) covers the software side of that skill set. For AI-focused credentials, see our ranking of [generative AI certifications](https://www.1stepgrow.com/articles/best-genai-certifications-2026). ## Frequently asked questions ### Which cloud certification should I get first? Whichever cloud your current or target employers actually run. If that is genuinely open, AWS Solutions Architect Associate has the largest job market and the deepest free study material. Market share is the tiebreaker, not the deciding factor. Check twenty local job postings for your target role before you book anything. ### Do cloud certifications increase salary? Associate-level certifications mostly affect whether you get shortlisted. Professional-level ones show up in offers, particularly when paired with matching experience. The certification alone, without the hands-on work behind it, does not survive a technical interview. Pair any certification with a deployed project you can explain in detail. ### How much does it really cost to get certified? Budget for more than the exam fee. An AWS associate exam costs $150, and two months of AWS Skill Builder at $29 a month adds $58. Practice tests and cloud lab usage add more, and renewal comes round every one to three years. Microsoft renewals are free, while Google charges $100 for its professional renewal exams. ### Are cloud certifications worth it if I have no experience? One associate certification plus one deployed, documented project is a strong entry combination. Three certifications and no project is a weak one — and the second is what most beginners accidentally build. Choose the associate exam for the cloud your target employers use, then spend the next month deploying something real. ### How long do cloud certifications stay valid? AWS certifications last three years. Microsoft role-based certifications expire after one year but renew free through a short online assessment. Google Cloud professional certifications last two years and associate ones three, while CKA and Terraform Associate each last two years. Renewal is a recurring commitment, so plan for it before you collect several certifications. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/top-cloud-certifications-2026/). Cite with the title, "1stepGrow" and a link._ --- # Digital Marketing Bootcamp With a Brand Audit: When to Pay for One > Is a digital marketing bootcamp with a live brand audit worth it? What the audit should cover, what free certifications offer, and what to ask before paying. - **Author:** Nishant Kiran — B2B SaaS & EdTech Content Strategist (https://www.1stepgrow.com/authors/nishant-kiran/) - **Published:** Aug 15, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Digital Marketing · **Format:** Guide · **Read time:** 8 min - **Canonical URL:** https://www.1stepgrow.com/digital-marketing-course-online/ ## Key takeaways - A brand audit deliverable is the strongest thing a marketing bootcamp can give you — if the brand is real and the audit is yours. - Ask whether the audit client is an actual business or a case study. The difference decides the portfolio value. - AI-integrated curricula are worth checking carefully: using the tools is table stakes, evaluating the output is the skill. - The bootcamp is buying you structure and a deliverable, not knowledge — the knowledge is free. A digital marketing bootcamp built around a live brand audit is worth paying for when two things are true: the audit is on a real business, and someone reviews your work in detail. If you reliably finish free courses, the channel knowledge itself is available at no cost from Google, HubSpot and others. Most digital marketing bootcamps sell the same channel modules: SEO, paid search, social, email, analytics and a capstone. Much of that is free: Google's Ads certifications cost nothing and HubSpot Academy's course library is free. What a free course cannot give you is a document with a real business's name on it and your reviewed recommendations inside, and a bootcamp that calls a case study "client work" does not give you that either. This guide is for career switchers and junior marketers deciding whether to pay. It shows what a serious brand audit covers, compares a bootcamp against free official training, explains how to test an "AI-integrated" claim, and lists five questions to put to any programme in writing. ## Why does the brand audit matter more than the certificate? Because marketing hiring runs on evidence, and a certificate is noise when everyone has one. A brand audit is evidence of exactly the things employers cannot assess from a CV: whether you can look at a messy real-world marketing operation, work out what is broken, prioritise the fixes by likely impact, and write it up so a busy person can act on it. That is the job: not running the campaign, but deciding which campaign is worth running. An audit you can walk an interviewer through, with the reasoning behind each recommendation and an honest note about what you could not determine, is worth more than any three channel certifications. ## What a brand audit in a digital marketing bootcamp should cover A serious audit reviews every channel the business uses, then ranks the fixes. Use this outline to judge a program's sample audit, or to run one yourself: | Audit section | What you examine | What a strong recommendation looks like | |---|---|---| | Search visibility | Indexing, titles, content gaps, local listings | A short list of pages to fix first, with the reason for each | | Paid search | Account structure, conversion tracking, wasted spend | A budget shift you can justify with the account's own numbers | | Social media | Posting consistency, engagement, audience fit | Which platform to drop, and which to double | | Content and email | What exists, what converts, what is missing | Two or three pieces worth producing next | | Measurement | Whether analytics and conversion tracking can be trusted | The tracking fix that must come before any other change | | Competitors | Who wins the searches and audiences the brand wants | One gap the brand can realistically close | | Priorities | Impact against effort for every recommendation | A ranked action plan a busy owner can follow | For the search section, Google's own [SEO Starter Guide](https://developers.google.com/search/docs/fundamentals/seo-starter-guide) is the right baseline; it says plainly that there are no secrets that will automatically rank a site first. Google also states that there are [no additional requirements or special optimisations](https://developers.google.com/search/docs/appearance/ai-features) needed to appear in AI Overviews or AI Mode, so be wary of any course selling a separate "AI search" trick. ## Is the audit client a real business or a case study? That single answer decides most of a bootcamp's portfolio value, so ask it before anything else. If real: you are producing consulting work, you will hit genuine constraints, you will have to present to someone who disagrees, and the result is portfolio-grade. This is the version worth paying for. If a case study: it is a class exercise with a nicer name. It is still useful practice, but do not pay a premium for it, and be careful how you present it. An interviewer who discovers your "client work" was a hypothetical will discount everything else you said. Ask directly, because a program confident in its model will tell you plainly, and often name the kinds of businesses involved. ## Digital marketing bootcamp or free certifications? The honest comparison is not bootcamp against nothing; it is bootcamp against the free official training. Here is what each route gives you: | | Free official training | Digital marketing bootcamp | |---|---|---| | Cost | Free | Paid, often substantially | | Channel knowledge | Strong, straight from the platforms | Similar content, repackaged | | Credential | Google Ads certifications, [valid for one year](https://support.google.com/google-ads/answer/9702955?hl=en) | A program certificate | | Structure and deadlines | None | Yes | | Feedback on your work | None | Yes, if mentors actually review it | | Portfolio deliverable | Only what you build yourself | A brand audit, if the client is real | As of September 2026, Google's Ads certifications need a score of at least 80% on an assessment, and [HubSpot Academy](https://academy.hubspot.com/courses) says its account and course library are free. Neither, however, gives you feedback on real work, which is the gap a good bootcamp fills. ## How can you test a bootcamp's AI-integration claim? Ask whether the course teaches you to verify AI output or only to produce it. Nearly every marketing programme now advertises AI integration, and the phrase covers an enormous range. **The weak version:** one module on prompting, some tool demonstrations, and a promise that you will be faster at producing content. **The strong version:** AI tooling used across the workflow, including research, drafting, analysis and reporting, with explicit attention to where it fails. Which claims need checking. Why the competitor analysis it generated is confidently wrong about two of the five competitors. How to tell whether the audience insight is real or merely plausible-sounding. The second version teaches judgement, which is what remains scarce. The first teaches a tool that will change in six months. ## What does a bootcamp fee actually buy? Structure, feedback, a deliverable and a cohort, but not the content. Be clear-eyed about this, because it changes whether the price is justified. - **Not the content.** Every channel topic in every bootcamp syllabus is available free, often taught better, from Google Skillshop, HubSpot Academy, Ahrefs' and Semrush's own material and a hundred good YouTube channels. - **Structure.** A schedule, deadlines and a cost to disappearing. For most people this is the real product. - **Feedback.** Someone reviewing your audit and telling you the recommendations are not prioritised. Free courses cannot do this. - **A deliverable.** The audit, if it is real. - **A cohort.** Peers, and sometimes an alumni network that produces referrals. If you reliably finish things you start, you are buying only the deliverable, and you could construct that yourself by auditing a local business for free. If you have abandoned two courses already, though, you are buying structure, feedback and a cohort, and they may genuinely be worth it. ## What should you ask before paying for a digital marketing bootcamp? Ask for evidence of the deliverable and the review behind it. Put these questions to the programme in writing: 1. **Can I see a sample audit from a previous cohort?** If they will not share one, that tells you something about the deliverable quality. 2. **Who reviews my work, and how much time do they spend on it?** "Industry mentor" can mean a detailed written review or a fifteen-minute group call. 3. **What is the audit client relationship?** A named business, a partner organisation, or a case study. 4. **What happens if I miss two weeks?** You will, so the answer matters. 5. **Does the AI module teach verification, or only production?** ## Common mistakes when choosing a marketing bootcamp - **Paying for content you could get free.** Check the free Google and HubSpot material first. - **Accepting a case study as client work.** It weakens your portfolio if an interviewer finds out. - **Ignoring certification expiry.** Google Ads certifications last a year, so a badge earned early may lapse before you job-hunt. - **Choosing by channel count.** Ten shallow modules teach less than one audit done properly. Whatever you decide, start auditing a real business in week one: a friend's shop, a local charity, your own employer. That habit is the thing that makes you employable, with or without the program. ## Where to go next Before any sales call, audit one local business yourself using the table above; it will show you what feedback you actually need to pay for. For the specialisation decision, see [SEO vs SMM vs SEM](https://www.1stepgrow.com/digital-marketing-specialization-course). For the measurement depth that separates senior marketers, read [advanced digital marketing and analytics courses](https://www.1stepgrow.com/articles/advanced-digital-marketing-course). For more practical guides, browse the [career guides hub](https://www.1stepgrow.com/career-guides). ## Frequently asked questions ### What is a brand audit in a marketing course? A structured review of a brand's current marketing: search visibility, paid performance, social presence, content, competitive position, and a prioritised set of recommendations. Done on a real business it is a genuine consulting deliverable; done on a case study it is a class exercise. ### Are digital marketing bootcamps worth the money? They are worth it when you need structure and a portfolio deliverable, and when the audit is on a real business. They are poor value if you reliably finish self-directed courses — the underlying content is freely available and frequently better. ### How long should a digital marketing bootcamp take? Three to six months part-time. Programs shorter than that cannot fit a meaningful audit; substantially longer ones are usually padding channel modules that could be covered in a week each. Check the schedule for how many weeks go to the audit itself, not only to channel lessons. ### Do employers care about bootcamp certificates in marketing? Not much. They care about what you can show — a campaign with numbers, an audit with recommendations, a site you ranked. The certificate gets you past a screening filter at best. Bring an audit or campaign you can walk an interviewer through, including what you would do differently next time. ### What does 'AI-integrated' mean in these courses? It varies enormously. At best it means using AI tools across research, production and analysis while learning to check their output. At worst it means one module on writing prompts. Ask which, specifically, before paying, and ask to see an example assignment where learners had to check AI output. --- _Source: 1stepGrow (https://www.1stepgrow.com/digital-marketing-course-online/). Cite with the title, "1stepGrow" and a link._ --- # Full Stack Developer Roadmap 2026: Six Stages With Exit Tests > A nine-month full stack developer roadmap: an exit test for every stage, free official docs, portfolio projects that stand out and how hiring really screens. - **Author:** Ayushi Kulshreshta — AI Engineer (https://www.1stepgrow.com/authors/ayushi-kulshreshta/) - **Published:** Aug 14, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Web Development · **Format:** Roadmap · **Read time:** 7 min - **Canonical URL:** https://www.1stepgrow.com/articles/full-stack-developer-roadmap-2026/ ## Key takeaways - Nine months of consistent evening study is realistic. Ninety days is not, whatever a bootcamp advertises. - Depth in one stack beats familiarity with four. Hiring managers test what you claim, not what you have heard of. - Every stage here ends in something deployed and public. Certificates are not checkpoints; working software is. - The hardest stage is four, where most people quit. It is also the one that separates candidates from hobbyists. This full stack developer roadmap gets you from a first line of HTML to a deployed application and a job-ready portfolio in about nine months at 12–15 hours a week. It has six stages, and each ends with an exit criterion that tells you when you are genuinely ready to move on. Most full stack roadmaps are a wall of logos that never tell you when you are done with a stage. That is how people spend a year "learning React" and still stall at the backend, the stage where most self-taught developers quit. Here, if you cannot meet a stage's exit criterion, you are not ready, however many hours you have logged. This roadmap is for career switchers and early developers studying around a full-time job. The timeline assumes twelve to fifteen hours a week, which is what work and a normal life actually leave. Adjust it proportionally, but do not compress it: the stages depend on each other in ways that punish skipping. - **Total duration:** 9 months - **Weekly commitment:** 12-15 hours - **Typical entry salary:** Varies by market — see the pay section ### Stage 1: Foundations — the web actually works like this (Weeks 1-6) Before frameworks, understand the platform. HTML semantics, CSS layout, and JavaScript without a library. Most people skip this and spend the next two years confused about why things break. **Skills:** Semantic HTML, CSS Flexbox & Grid, Responsive design, JavaScript fundamentals, DOM & events, Git basics **Projects:** - A responsive multi-page site built without any framework - A browser tool — unit converter, budget calculator — using vanilla JS only **Outcome:** You can build and style a working page from a blank file without reaching for a tutorial, and explain what the browser is doing at each step. ### Stage 2: JavaScript in depth and the modern toolchain (Weeks 7-12) The language properly: async, modules, closures, error handling. Then the tooling every job posting assumes you already know — package management, bundlers, TypeScript. **Skills:** ES2023+ syntax, Promises & async/await, Fetch and REST, npm & modules, TypeScript basics, Debugging in DevTools **Projects:** - A weather or news dashboard consuming a public API with real error and loading states - Convert an earlier project to TypeScript **Outcome:** You can consume any documented API, handle failure states properly, and read a TypeScript error without panicking. ### Stage 3: Frontend framework — pick one and go deep (Weeks 13-20) React remains the safest default: it is the most-used front-end library among professional developers. Learn component design, state management and data fetching properly rather than collecting three frameworks at surface level. **Skills:** React & hooks, Component architecture, Client & server state, Routing, Forms & validation, Accessibility basics **Projects:** - A multi-view application with authentication-gated routes - One component library piece built to be genuinely reusable, with documented props **Outcome:** You can build a multi-screen application from a blank project, and explain your state management choices to someone who disagrees with them. ### Stage 4: Backend, databases and the part where people quit (Weeks 21-30) The hardest stage, and the one that separates people who can build features from people who can build systems. Servers, data modelling, authentication, and the security basics you cannot skip. **Skills:** Node.js & Express (or your language of choice), REST API design, PostgreSQL & schema design, Authentication & sessions, Password hashing & OWASP basics, Environment configuration **Projects:** - A full CRUD API with authentication, validation and meaningful error responses - A relational schema for a non-trivial domain, with the migration history intact **Outcome:** You can design a schema, expose it safely through an API, and explain why your auth implementation is not the naive version. ### Stage 5: Full stack integration — ship the thing (Weeks 31-36) Join the halves and put it in front of real users. This project is the one your interviews will be about, so choose a problem you actually care about. **Skills:** Frontend-backend integration, Deployment & CI, Environment management, Basic Docker, Monitoring & logging, Performance basics **Projects:** - One complete application, deployed, with a custom domain and a README that explains the trade-offs - A CI pipeline that runs tests on every push **Outcome:** You have a live URL you would show a hiring manager, and you can walk through every architectural decision and its alternative. ### Stage 6: Job readiness — the part nobody schedules (Weeks 37-40) Interviewing is a separate skill from building. Treat it as its own stage rather than something you improvise after applying. **Skills:** Data structures for interviews, System design fundamentals, Explaining your project out loud, Reading a codebase quickly, Salary negotiation **Projects:** - One meaningful pull request merged into an open source project - A written case study of your main project, published **Outcome:** You can explain your project under questioning, read unfamiliar code in an interview, and negotiate without flinching. ## How long does each stage of the full stack developer roadmap take? The full plan is 40 weeks. At 12–15 hours a week that is roughly 480–600 hours, and the per-stage hours below are simply weeks multiplied by that range. | Stage | Weeks | Approximate hours | Exit criterion in one line | |---|---|---|---| | 1. Web foundations | 1–6 | 72–90 | A styled page from a blank file, no tutorial | | 2. JavaScript and tooling | 7–12 | 72–90 | Any documented API consumed, failure states handled | | 3. Frontend framework | 13–20 | 96–120 | A multi-screen app with defensible state choices | | 4. Backend and databases | 21–30 | 120–150 | A safe, authenticated API over a designed schema | | 5. Integration and deployment | 31–36 | 72–90 | A live URL with CI and a README of trade-offs | | 6. Job readiness | 37–40 | 48–60 | You can defend your project under questioning | ## Which official docs should you learn from at each stage? Official documentation is free, current and what working developers actually use. As of September 2026, these are the best starting points: - **Stages 1–2:** MDN's [Learn web development](https://developer.mozilla.org/en-US/docs/Learn_web_development) curriculum, which covers HTML, CSS layout, JavaScript, accessibility and version control, followed by the [TypeScript handbook](https://www.typescriptlang.org/docs/handbook/intro.html). - **Stage 3:** the official [React tutorial and docs](https://react.dev/learn), which teach the concepts you will use daily. - **Stage 4:** the [Node.js learning pages](https://nodejs.org/learn), the [PostgreSQL tutorial](https://www.postgresql.org/docs/current/tutorial.html) and OWASP's [Password Storage Cheat Sheet](https://cheatsheetseries.owasp.org/cheatsheets/Password_Storage_Cheat_Sheet.html), which recommends Argon2id for hashing passwords. - **Stage 5:** Docker's [getting started guide](https://docs.docker.com/get-started/) and the [GitHub Actions documentation](https://docs.github.com/en/actions) for CI. If you have never set up version control, start with [installing Git and sharing files with GitHub](https://www.1stepgrow.com/articles/git-and-github-tutorial). These choices are not arbitrary. In the [Stack Overflow Developer Survey 2025](https://survey.stackoverflow.co/2025/technology), professional developers reported extensive work with Node.js (49.1%), React (46.9%) and PostgreSQL (58.2%), while 71.1% of all respondents used Docker. ## Why do people quit at stage four? Backend work gives little visible feedback, so it feels like no progress. Stage four is also the longest in the timeline, which is deliberate and the honest part of this roadmap. Frontend work has immediate visual feedback. You change something, you see it, and it feels like progress. Backend work does not. You spend an evening on an authentication flow and at the end you have... a login that works, which is exactly what you had conceptually before you started. This is where the dropout happens. It is also why backend competence is a hiring differentiator: many self-taught developers never get through it, so the ones who do stand out. If you are in stage four and it feels like a slog, that is not a sign you chose wrong. It is the stage doing its job. ## Which portfolio projects should you build? Build something with a real user and a real constraint, not a tutorial clone. Everyone says "build projects"; the useful advice is about which ones. **Avoid:** todo apps, weather apps as your main project, e-commerce clones, anything from a tutorial you followed to completion. Not because they are bad exercises — several appear in the stages above — but because as a headline portfolio project they signal that you can follow instructions. **Build instead:** something with a real constraint. For example: - An application for a hobby you have, such as bookings for a club or kit tracking for a team. - A tool for a problem at your current job, like an internal request tracker with roles and audit history. - Something for a local community group, where real users will report real bugs. The constraint is what generates interesting decisions, and interesting decisions are what you talk about in interviews. The best portfolio project any mentee of mine has shipped was a scheduling tool for their badminton club. Forty users. Genuinely awkward requirements. It got them four interviews because they could talk for an hour about why the recurring-booking logic was harder than it looked. ## How does hiring screen full stack developers? Hiring for junior full stack roles usually tests whether your claims survive contact with questions. A typical process includes: - **CV and portfolio review.** A live link, a readable README and a clean commit history do more than a skills list. - **Take-home task or live coding.** Often a small feature across the front end and an API, judged on structure, error handling and tests. - **Project deep-dive.** Why this schema, why this state library, and what happens when two users edit the same record. - **Fundamentals.** HTTP, authentication, SQL joins and basic data structures. - **Light system design.** For juniors, usually "how would this app handle ten times the users?" rather than a full distributed design. ## What does a full stack developer earn? Full stack development has no occupation code of its own, so official pay data comes from the closest categories. The survey rows are self-reported and are best read as ratios rather than exact levels. | Source | Population | Median annual pay | |---|---|---| | [US BLS, May 2025](https://www.bls.gov/ooh/computer-and-information-technology/web-developers.htm) | US web developers | $92,650 | | [US BLS, May 2025](https://www.bls.gov/ooh/computer-and-information-technology/software-developers.htm) | US software developers | $135,980 | | [Stack Overflow Developer Survey 2025](https://survey.stackoverflow.co/2025/work) | Full-stack developers, global | $72,509 | | Stack Overflow Developer Survey 2025 | Full-stack developers, US | $138,000 | | Stack Overflow Developer Survey 2025 | Full-stack developers, India | $13,949 | Full-stack was also the most common role in that survey, chosen by 27% of all respondents. No official Indian source publishes a web developer salary band, so treat any single "entry salary" figure you see for India with caution. ## Common mistakes on the way to junior-ready These are the errors that most often add months to this path: 1. **Starting with a framework.** Without stage one, every React bug looks like magic. 2. **Rolling your own authentication carelessly.** Store passwords with a recommended hashing algorithm, and never in plain text or with a fast general-purpose hash. 3. **Skipping SQL for an ORM.** You will still need to read the queries it generates when something is slow. 4. **Never deploying.** An app that only runs on your laptop hides the configuration, environment and security problems employers care about. 5. **Stopping after a burst.** Consistency matters far more than total hours, as the next section explains. ## How long does it really take to become a full stack developer? Nine months of consistent work gets most people to junior-ready. Some do it in six. Plenty take fourteen, particularly with a demanding job or caring responsibilities, and that is completely normal. What matters far more than the total is consistency. Ten hours a week every week beats thirty hours in a burst followed by three weeks off — and by a much wider margin than people expect, because the restart cost after a gap is brutal. ## Where to go next Start stage one with MDN's curriculum tonight, and write down its exit criterion where you will see it. If you are weighing web development against other tracks, our [Data Science vs Machine Learning vs AI](https://www.1stepgrow.com/articles/data-science-vs-machine-learning-vs-ai) comparison covers the adjacent options, and [data science vs full-stack development](https://www.1stepgrow.com/articles/data-science-vs-full-stack-development) compares the two directly. For the broader question of turning study into income, see [From Learning to Earning](https://www.1stepgrow.com/articles/learning-to-earning-tech-career-roadmap). ## Frequently asked questions ### Can I become a full stack developer in three months? Not from zero, and anyone advertising it is selling completion rather than employment. Three months gets you to roughly the end of stage two. The people who appear to do it in three months almost always had prior programming experience, often in another language or an adjacent technical job. ### Should I learn React, Vue or Angular? React, unless a specific employer you are targeting uses something else. In the Stack Overflow Developer Survey 2025, 46.9% of professional developers reported extensive work with React, against 19.8% for Angular and 18.4% for Vue.js. The concepts transfer, so learn one properly rather than three superficially. ### Do I need a computer science degree? No. It helps in large-company hiring pipelines and for roles with algorithmic depth. For product engineering, a deployed portfolio and the ability to reason about your own code carry more weight in almost every process we examined. The US Bureau of Labor Statistics still lists a bachelor's degree as typical entry-level education, so expect some filters. ### Is full stack development still a good career with AI writing code? The floor rose — trivial CRUD work is genuinely being automated. The ceiling did not move. Engineers who understand systems, data modelling and trade-offs remain in demand, because someone has to review and integrate what the tools produce, and to own the result when it fails in production. ### How many projects do I need in my portfolio? Three, with deliberately different depth. One substantial application you can discuss for forty minutes, one smaller focused piece, and one open source contribution. Ten tutorial clones are worse than one real thing, because they show you can follow instructions rather than make decisions. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/full-stack-developer-roadmap-2026/). Cite with the title, "1stepGrow" and a link._ --- # What Is Data Science? A Decision Discipline, Explained > What is data science? A plain-English guide to what data scientists do, the project workflow, the skills that matter and why organisations still pay for it. - **Author:** Nishant Kiran — B2B SaaS & EdTech Content Strategist (https://www.1stepgrow.com/authors/nishant-kiran/) - **Published:** Aug 13, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Guide · **Read time:** 8 min - **Canonical URL:** https://www.1stepgrow.com/articles/what-is-data-science/ ## Key takeaways - Data science is a decision discipline, not a modelling discipline — the model is often the smallest part. - Getting and cleaning data usually takes longer than the analysis, and communicating the result takes longer than most syllabi admit. - The value is in reducing uncertainty about a decision someone is already making. - Better tooling has raised the floor, not removed the job — someone still has to ask the right question. What is data science? Put simply, it is the work of using data, statistics and code to reduce uncertainty about a decision someone has to make. In practice that means framing a question, finding and cleaning the data, analysing it, checking whether the result holds up, and explaining it to the person who will act. Most definitions stop at the Venn diagram of statistics, programming and domain knowledge, which is technically true and tells you nothing about the job. The decision-first framing explains what is otherwise confusing: why so much of the work is unglamorous data cleaning, why communication is a graded interview skill, and why the fanciest model frequently loses to a simple one. This overview is for students, career changers and managers who want to know what the work involves before investing in it. It covers the official duty list, a project from question to memo with a worked churn example, how data science differs from its neighbours, the skills in priority order, and where to start. ## What is data science, according to official job data? Official job data describes a role built mostly around data, judgement and communication, and it is more concrete than most textbook definitions. The US Bureau of Labor Statistics lists five core duties for [data scientists](https://www.bls.gov/ooh/math/data-scientists.htm): - deciding which data are available and useful for a project, before any analysis starts - collecting, categorising and analysing data - creating, validating, testing and updating algorithms and models - presenting findings with data visualisation software - making business recommendations to stakeholders Only one of the five is about models. The rest are about data, judgement and communication, and that balance also matches what the job feels like from the inside. ## What does a data science project look like? A realistic project runs roughly like this. 1. **Someone asks a badly formed question.** "Why is churn up?" or "Should we open in Pune?" The first real task is turning that into something answerable, which usually means going back and negotiating what would count as an answer. 2. **You find out what data exists.** There is almost always less than you hoped, in a worse state than you were told, with a definition that changed in 2023 and nobody documented. 3. **You clean it.** On most projects this takes more time than any other step, because real data is messy. It is also where the important discoveries happen: the reason the numbers look strange is frequently the answer to the question. 4. **You analyse.** Sometimes that means a model, but not always. More often it means a well-constructed comparison, a cohort breakdown or a properly designed experiment. The sophistication should match the question, not your résumé. 5. **You check whether you believe it.** Is the effect real or noise? What else could explain it? What would change your mind? This step, above all, separates data science from making charts. 6. **You communicate.** A memo, a deck, a conversation. If the decision-maker cannot understand it, act on it or disagree with it for good reasons, the analysis might as well not have happened. That balance is especially uncomfortable for people who came for the modelling. Getting and cleaning data usually takes longer than the analysis, and communication takes longer than most syllabi admit. ### A worked example: the churn question Take "Why is churn up?" at a subscription business. First, you agree what churn means — cancelled within the month, or simply not renewed? Next, you break cancellations down by signup month and find that the spike comes almost entirely from customers who joined during a discount campaign. Then you check whether customers from earlier campaigns churned faster too. Suppose they did. In that case the answer is not "the product got worse"; instead, it is "discount cohorts churn faster, and we just ran a big one". No machine learning was involved, yet the finding changes next quarter's marketing plan. That, in miniature, is the answer to what is data science: a better decision, backed by evidence someone can check. ## How is data science different from analytics, statistics and machine learning? The neighbouring fields overlap heavily, so the clearest difference is the question each one starts from. | Field | Starting question | Typical output | |---|---|---| | Data analytics | What happened, and by how much? | Dashboards and reports | | Statistics | How confident can we be in this conclusion? | Estimates, tests and study designs | | Machine learning | Can a system learn to predict this automatically? | Trained models running in production | | Data science | What should we decide, given the data? | Recommendations backed by analysis, sometimes a model | The labels blur in real job ads. For example, O*NET, the US Department of Labor's occupation database, files [business intelligence analysts](https://www.onetonline.org/link/summary/15-2051.01) as a specialism under the data scientist occupation. So read the responsibilities rather than the title. The most common confusion gets its own guide: [data analyst vs data scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist). ## Why do organisations pay for data science? There are three reasons, in rough order of how much they are worth. **Decisions under uncertainty.** Should we change the price? Is the new onboarding flow better? These carry real money, yet whoever argues most confidently often decides them. A good data scientist replaces confidence with evidence. **Scale.** A human can review a hundred loan applications, whereas a model can review a hundred thousand, consistently, with a measurable error rate. Consistency is often the bigger win. **Discovery.** Sometimes the data contains something nobody suspected — a segment behaving differently, or a process quietly failing. This is the least predictable value, but occasionally the largest. Official figures reflect that demand. As of September 2026, the latest BLS data put the median data scientist salary at $120,230 (May 2025) and project employment to grow 35% from 2025 to 2035, much faster than the average for all occupations. The largest employers include publishing and broadcasting, insurance carriers and computer systems design firms, so the work is not confined to tech companies. ## Which data science skills matter most? Here they are, honestly ordered. **SQL** matters more than anything else on this list, and weak SQL is the most common reason capable candidates fail early interview rounds. **Statistics** — distributions, inference and experiment design — comes next, because the ability to say whether a result is real is the core of the discipline. **Programming** in Python, mostly for pandas and modelling, needs to be competent rather than brilliant. Both SQL and Python are mainstream well beyond data work: in the [2025 Stack Overflow Developer Survey](https://survey.stackoverflow.co/2025/technology), 58.6% of respondents had done extensive work in SQL over the past year and 57.9% in Python. **Communication** is graded explicitly in every interview process worth joining. Analyses that nobody acts on have no value, whatever their technical quality. **Domain knowledge** is underrated and hard to teach. Knowing which numbers are unreliable and what a plausible effect size looks like in your industry takes years — which is why career switchers with deep industry backgrounds often outperform fresh graduates. ## Has AI made data science obsolete? No. The tooling has genuinely improved: AutoML fits models, notebooks can draft code, and a competent assistant can produce a passable exploratory analysis. What has not changed is that someone must decide which question is worth asking, whether the data can answer it, whether the answer is trustworthy, and what the organisation should do about it. Those are judgement tasks, and they were always the hardest parts anyway. The realistic effect is a raised floor. For example, work that consisted of producing routine reports is shrinking. Meanwhile, work that consists of framing problems and defending conclusions is in more demand, because there is now far more output that needs someone to evaluate it. ## Where should you start? If this sounds like work you would enjoy, the sequence that works is: 1. SQL first. 2. Statistics alongside it, rather than after. 3. Python after that. 4. Modelling last, once the foundations hold. Our [data scientist roadmap](https://www.1stepgrow.com/articles/data-scientist-roadmap-2026) then lays that out as a stage-gated path with exit criteria. If you are not sure this is the right field, [AI vs machine learning vs deep learning](https://www.1stepgrow.com/articles/ai-vs-machine-learning-vs-deep-learning) explains how the adjacent fields fit together. Who should think twice? If you dislike ambiguity, or you would rather build systems than argue for decisions, then software or machine learning engineering may suit you better than data science. ## Related reading [Tools and techniques for data science](https://www.1stepgrow.com/tools-and-techniques-for-data-science) covers the practical stack, such as SQL, pandas and scikit-learn. [The role of programming in data science](https://www.1stepgrow.com/articles/programming-for-data-science) addresses how much code you actually need. ## Frequently asked questions ### What exactly does a data scientist do? Takes a business question that is usually badly defined, works out what data could answer it, checks whether the answer is trustworthy, and communicates it to someone who will act. Modelling is part of that, but a smaller part than most course syllabi imply. ### Is data science just statistics with a better job title? Statistics is the intellectual core, but the job adds programming, data engineering and communication. A statistician who cannot get their own data or explain the result to a sceptical executive will struggle in most data science roles. ### Do I need a PhD to work in data science? No. Research-heavy teams and some large enterprises still prefer advanced degrees, but most industry roles care far more about whether you can get to a defensible answer and explain it. BLS lists a bachelor's degree as the typical entry-level education for data scientists in the US, and a strong portfolio often matters more than the degree itself. ### Is data science still worth entering? Yes, with a caveat. The routine reporting end is shrinking as tooling improves, while the experimentation, causal inference and decision-support end is growing. BLS projects US data scientist employment to grow 35% from 2025 to 2035. Aim at the judgement-heavy end of the field, where framing problems and defending conclusions matter more than producing reports. ### What is the difference between data science and data analytics? Analytics leans toward describing what happened; data science leans toward explaining why and predicting what will. In practice the boundary is blurry and the titles are used inconsistently — read the responsibilities, not the label. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/what-is-data-science/). Cite with the title, "1stepGrow" and a link._ --- # Data Science Tools and Techniques: The Stack That Actually Gets Used > The data science tools working analysts actually use, what each is for, when to learn it, and the validation techniques that matter more than any library. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Aug 12, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/tools-and-techniques-for-data-science/ ## Key takeaways - SQL, Python, pandas, scikit-learn, one plotting library and Git cover the overwhelming majority of real work. - Techniques outlast tools. Cross-validation and experiment design will still matter when today's libraries are gone. - Tool sprawl on a CV signals shallowness, not breadth — interviewers test what you list. - The most valuable technique nobody teaches is honest validation. The data science tools that do most of the real work are SQL, Python, pandas, scikit-learn, one plotting library, Git, a notebook plus an editor, and one dashboard tool. Learn those properly before anything else. Techniques such as honest validation and experiment design matter more than any addition to that list. Open a typical data science syllabus and you will find dozens of tools; the stack a working data scientist leans on is closer to eight. Syllabi are written to look comprehensive, while jobs get done with whatever is sufficient. This guide is for students and junior analysts deciding what to learn next. It covers what each of the eight tools is for and when to learn it, the six techniques interviews actually probe, why a long tool list weakens a CV, and who should ignore this advice. ## Which data science tools do you actually need? Eight tools cover the overwhelming majority of day-to-day work. They are listed here roughly in order of how much you will use them. ### Getting and shaping the data **SQL.** It is the most-used tool in the field by a wide margin, and also the one candidates most consistently under-prepare. Every project starts with getting the data, and the data is nearly always in a database. Window functions and CTEs are not advanced — they are Tuesday. **Python.** This is the general-purpose glue. You need competence, not virtuosity: functions, data structures, control flow, and enough comfort to read someone else's code. **pandas.** Most analysis actually happens here: `groupby`, merges, reshaping and missing-value handling. Expect to spend more time here than anywhere else, and expect it to be occasionally infuriating. Our [NumPy series](https://www.1stepgrow.com/articles/numpy-tutorial) covers the array layer underneath it. ### Visualising, modelling and sharing the work **One plotting library.** Use matplotlib for control, seaborn for speed, or [plotly](https://www.1stepgrow.com/articles/plotly-tutorial) for interactivity. Learn one properly, because switching later is easy, whereas knowing three badly helps nobody. **scikit-learn.** This is the default modelling toolkit for everything that is not deep learning — which, in business data science, is most things. It has a consistent API, excellent documentation and sensible defaults. **Git.** It is not optional any more. If your analysis cannot be reproduced or reviewed, it is a liability. Our [Git and GitHub guide](https://www.1stepgrow.com/articles/git-and-github-tutorial) covers the basics. **A notebook and an editor.** Use Jupyter for exploration and a real editor for anything that will run twice. The failure mode is doing everything in notebooks and ending up with code nobody can run, including you. **A dashboard tool.** Power BI, Tableau or Looker, depending on the employer. Learn the one your organisation uses; none of them is hard. These are not niche choices, either. In the [2025 Stack Overflow Developer Survey](https://survey.stackoverflow.co/2025/technology), still the latest edition as of September 2026, 58.6% of all respondents had done extensive work in SQL over the past year and 57.9% in Python, and the survey reported a 7 percentage point rise in Python use from 2024 to 2025. Those figures cover developers in general rather than data scientists alone, yet they show how widely transferable the core stack is. ## Data science tools at a glance | Tool | What it is for | When to learn it | Common alternatives | |---|---|---|---| | SQL | Getting and aggregating data from databases | First, before anything else | None worth substituting | | Python | Glue code, automation, analysis | Straight after basic SQL | R, for academia and biostatistics | | pandas | Cleaning and reshaping tabular data | As soon as you know Python basics | Polars, for larger or faster workloads | | matplotlib, seaborn or plotly | Charts for analysis and reports | Alongside pandas | Your BI tool's built-in charts | | scikit-learn | Classical machine learning and validation | After statistics fundamentals | statsmodels, for inference | | Git | Version control and code review | Before your first shared project | None in practice | | Jupyter plus an editor | Exploration, then maintainable code | From day one | VS Code notebooks | | Power BI, Tableau or Looker | Dashboards for non-technical users | When a role or employer requires it | Excel for small audiences | Everything else — Spark, Airflow, dbt, MLflow, cloud platforms — is genuinely valuable *when a role requires it* and premature otherwise. ## Why do techniques outlast tools? The libraries in that list will eventually be replaced. The techniques below will not, and they are also what interviews actually probe. **Exploratory data analysis, done sceptically.** Good EDA is not producing twenty plots. Instead, it means asking what shape the data has, what is missing, what looks impossible, and who would know why. The most valuable finding in a project is frequently that a column means something different from what its name suggests. **Cross-validation and honest holdouts.** Know why a single train-test split can mislead you. As the scikit-learn [cross-validation guide](https://scikit-learn.org/stable/modules/cross_validation.html) explains, tuning settings until the test score looks good lets knowledge of the test set leak into the model. Time-ordered data needs a time-ordered split for the same reason, since standard k-fold assumes independent samples. This is where most self-taught practitioners have a real gap. **Leakage detection.** This is the most common way to produce a model that looks brilliant and is worthless. scikit-learn's [common pitfalls page](https://scikit-learn.org/stable/common_pitfalls.html) defines leakage as using information that would not be available at prediction time. Spotting that is a skill, not a checklist. **Metric selection.** Accuracy is almost always the wrong metric. For example, a fraud model that flags nothing can score 99% accuracy when only 1% of transactions are fraudulent. Choosing a metric that reflects the real cost of each error type is a judgement call, and being able to defend it is a strong interview signal. **Experiment design.** Sample size, power, duration, and why stopping a test early because it looks good invalidates it. Most business questions are causal, yet most people answer them with correlational analysis. **Error analysis.** Look at what your model got wrong and find the pattern. It is consistently more valuable than another round of hyperparameter tuning, and consistently skipped. ## Why does tool sprawl hurt your CV? There is a temptation to list everything you have touched. Resist it. A CV listing fifteen technologies invites an interviewer to pick the two you know least. A CV listing six, all of which you can discuss in depth, produces a much better conversation. The same applies to learning. Depth in pandas and scikit-learn will serve you better than surface familiarity with pandas, Spark, Dask, Polars and Ray. Besides, when you do need one of those, the fundamentals transfer in days. ## What should you learn next? It depends on where you are starting from. - **If you are starting out:** SQL, then Python, then pandas, then scikit-learn. Keep statistics running alongside throughout. - **If you already work with data:** the gap is usually validation discipline rather than tooling. Learn cross-validation properly, learn to detect leakage, and learn to design an experiment. Those three will improve your work more than any new library. - **If a job ad demands a big-data stack:** learn the specific tool it names, such as Spark or dbt, only once the core above is solid. Who should ignore this advice? Data engineers and ML engineers, mainly. For those roles, orchestration, cloud and distributed tools are the job itself rather than an optional extra. ## Related reading [What data science is and why it matters](https://www.1stepgrow.com/articles/what-is-data-science) covers the job itself. [The role of programming in data science](https://www.1stepgrow.com/articles/programming-for-data-science) addresses how much code you genuinely need, and [Data Analyst vs Data Scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist) compares the two roles that share this stack. ## Frequently asked questions ### Which programming language should I learn for data science? Python, unless you are heading for academia or biostatistics, where R remains strong. Python has more industry roles and integrates better with engineering workflows. It is also widely used beyond data work: 57.9% of respondents to the 2025 Stack Overflow Developer Survey had worked extensively in it over the past year. Learn SQL alongside it whichever language you choose. ### Do I need to learn Spark or big data tools? Only if you will actually work with data that does not fit in memory, which is a smaller share of jobs than course syllabi suggest. Most business datasets fit comfortably on a laptop. Learn it when a role requires it. ### Is Excel still relevant? Yes, and dismissing it is a mistake juniors make. It is how most of the organisation looks at data, and being able to hand someone a well-built spreadsheet is often more useful than a dashboard nobody opens. ### How many tools should I list on my CV? Only the ones you could be interviewed on. A short list you genuinely know beats a long one where two questions expose the gaps. As a rule of thumb, six tools you can discuss in depth — say SQL, Python, pandas, scikit-learn, one plotting library and Git — make a stronger CV than fifteen you have merely touched. ### What about AI coding assistants? Use them — they are genuinely good at boilerplate and transformation code. Be careful with anything statistical: they will produce confident, plausible and wrong analysis code, and you need enough knowledge to catch it. --- _Source: 1stepGrow (https://www.1stepgrow.com/tools-and-techniques-for-data-science/). Cite with the title, "1stepGrow" and a link._ --- # Programming for Data Science: How Much Code Do You Really Need? > Programming for data science: how much SQL and Python you need, what reproducible code looks like, the notebook traps, and where AI assistants help or hurt. - **Author:** Ayushi Kulshreshta — AI Engineer (https://www.1stepgrow.com/authors/ayushi-kulshreshta/) - **Published:** Aug 11, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/programming-for-data-science/ ## Key takeaways - Data science needs competent programming, not software engineering excellence — but the bar is higher than most beginners assume. - The line is reproducibility: if your analysis cannot be rerun by someone else, it is not finished. - SQL is programming too, and it is the language you will use most. - Notebook-only workflows are the most common source of unreproducible, unreviewable work. - In the 2025 Stack Overflow Developer Survey, 66% of respondents named AI output that is almost right, but not quite, as their top frustration with AI tools. Programming for data science needs to be competent, not expert. You must write SQL and Python that runs twice: code someone else can rerun to get your numbers, read without you in the room, and defend in an interview. You do not need software architecture, design patterns or systems programming. The bar is easy to misjudge in both directions, and both mistakes cost months. Overshoot and you study patterns no analysis role uses. Undershoot and you share a notebook that only works if its cells run in the order you once clicked them. This guide is for career changers and early-career analysts asking how good at coding they need to be. It defines the bar level by level, shows what reproducible code looks like, and explains where AI assistants help and where they quietly introduce errors such as data leakage. ## How much programming for data science do you need? **You need to write code that runs twice.** That sounds trivial, yet it is the whole thing. It means someone else can take your project, run it, and get your numbers. That in turn requires paths that are not hardcoded to your desktop, a stated environment, a fixed random seed, cells that execute top to bottom, and no manual step you forgot to write down. Most analysis code fails this test. Passing it consistently puts you ahead of a surprising share of working data scientists. **You do not need** design patterns, deep inheritance hierarchies, microservice architecture, or the ability to optimise at the assembly level. Those are software engineering skills — adjacent and occasionally useful, but not your job. The table below shows how that bar rises as you progress. | Level | What you can do | Sign you have reached it | |---|---|---| | Beginner | Follow a tutorial, run notebook cells, write simple queries | You can reproduce an example but not adapt it | | Job-ready | Write joins, window functions, pandas transformations and small functions unaided | A colleague can rerun your project from a clean environment | | Strong | Package reusable code into modules, test key assumptions, use Git daily | Your analysis survives a code review without a rewrite | | Beyond the role | Design services, optimise performance, own production systems | You are doing ML or software engineering, not data science | ## What code does a data scientist actually write? **SQL, constantly.** It is programming that people forget to call programming: joins, aggregations, window functions and CTEs. You will write more SQL than Python in most roles, and it is what interviews test first. **Data transformation.** In pandas that means loading, filtering, `groupby`, merging, reshaping and handling missing values. This is the bulk of the work. **Analysis and modelling.** Usually you assemble library calls rather than implement algorithms. Knowing which function to use, and why, matters far more than being able to write the algorithm from scratch. **Small utilities.** These are functions you write once and reuse, such as a loader, a cleaner or a plotting helper. This is where the step up from beginner to competent happens. **Occasionally, production code.** Wrapping a model in an API or writing a scheduled job. Many teams hand this to engineers; even so, being able to do it yourself makes you considerably more useful. ## Why do notebooks cause so many problems? The single most common code-quality failure in data science is not bad algorithms. Instead, it is the notebook that only works if you run the cells in the order you happened to run them three weeks ago. You know the symptoms: a variable defined in a cell you later deleted, cell 14 run before cell 9, a file path pointing to a folder that no longer exists, and a result you cannot reproduce but are fairly sure was right. The fix is cheap and habitual. 1. **Restart and run everything.** Restart the kernel and run all cells, top to bottom, before you share anything. If it fails, it was already broken — you just had not noticed. 2. **Move reusable code into modules.** Anything you use more than twice belongs in a `.py` file that you import. Notebooks are for exploring; modules are for things that must keep working. 3. **Pin the environment.** Create a virtual environment per project with Python's built-in [venv module](https://docs.python.org/3/library/venv.html), and record the package versions you used. 4. **Keep paths relative and configuration at the top.** Then the project runs on any machine, not just yours. 5. **Commit to Git.** Our [Git and GitHub guide](https://www.1stepgrow.com/articles/git-and-github-tutorial) covers enough to be useful in an afternoon. ## What does reproducible code look like? Here is the idea in miniature. The function fixes its random seed, and the assertions check both an assumption about the data and the reproducibility itself. ```python import random def sample_customers(customer_ids, k, seed=42): """Draw the same random sample every time the analysis is rerun.""" rng = random.Random(seed) return sorted(rng.sample(customer_ids, k)) customer_ids = list(range(1000, 1100)) assert len(set(customer_ids)) == len(customer_ids), "customer IDs are not unique" first_run = sample_customers(customer_ids, 5) second_run = sample_customers(customer_ids, 5) assert first_run == second_run, "sample is not reproducible" print(first_run) ``` Output: ```text [1003, 1014, 1035, 1081, 1094] ``` Create the generator with `random.Random()` instead, with no seed, and the two runs will almost certainly differ, so the second assertion fails. The same principle applies in libraries: scikit-learn's guide to [controlling randomness](https://scikit-learn.org/stable/common_pitfalls.html) explains how the `random_state` parameter decides whether results repeat from one run to the next. ## Which habits mark a competent analyst-programmer? **Functions over copy-paste.** The third time you paste the same transformation, make it a function. It is the single highest-return habit in analysis code. **Meaningful names.** `df2` and `temp` are how you lose an hour next month, whereas `orders_by_region` costs nothing. **Assertions on your assumptions.** `assert df['user_id'].is_unique` catches a whole category of silent errors that otherwise surface as a wrong number in a board deck. **Handle failure explicitly.** Data pipelines break because the source changed. Our [exception handling guide](https://www.1stepgrow.com/articles/python-exception-handling) covers doing this without swallowing the errors you needed to see. **Comment the why.** Not what the line does — the reader can see that. Explain why the filter excludes 2023, or why this column is cast to string. ## Where do AI coding assistants fit? They help most with the tedious parts: boilerplate, transformations, plotting syntax you can never remember, and docstrings. Adoption is now mainstream. In the [2025 Stack Overflow Developer Survey](https://survey.stackoverflow.co/2025/ai), 84% of respondents were using or planning to use AI tools in their development process. However, they are dangerous at the statistical parts. They will write a train-test split that leaks, choose accuracy on imbalanced data, or implement a time series validation that peeks at the future — and it will all look confident and idiomatic. The same survey found that the top frustration, cited by 66% of respondents, was AI output that is "almost right, but not quite". Leakage is the classic example. scikit-learn defines [data leakage](https://scikit-learn.org/stable/common_pitfalls.html) as using information at training time that would not be available at prediction time, which produces overly optimistic scores. An assistant will not reliably warn you about it, so you have to recognise it yourself. Therefore, the floor for what you must understand has gone *up*, not down. You can write more code with less recall, but you need more judgement to review it. ## Who can get by with less code? Not everyone needs this full bar. If you are aiming at reporting or BI analyst roles, excellent SQL, a dashboard tool and light Python will usually carry you. On the other hand, if you want to move towards ML engineering later, aim above the "strong" row in the table, because testing, packaging and deployment become the job. ## Related reading Start with our Python fundamentals series: [building blocks](https://www.1stepgrow.com/articles/python-basics), [functions](https://www.1stepgrow.com/articles/python-functions) and [NumPy](https://www.1stepgrow.com/articles/numpy-tutorial). For the wider picture, see [what data science is and why it matters](https://www.1stepgrow.com/articles/what-is-data-science), and for the rest of the stack read [tools and techniques for data science](https://www.1stepgrow.com/tools-and-techniques-for-data-science). ## Frequently asked questions ### Do data scientists need to be good programmers? Competent, not excellent. You need code that runs reliably, that a colleague can read, and that you can explain in an interview. You do not need design patterns, deep object-oriented architecture or systems programming. The practical test is reproducibility: someone else can rerun your project from a clean environment and get the same numbers. ### Can I do data science without coding at all? There are no-code tools, and they are genuinely useful for reporting. But you will hit a ceiling quickly — non-standard transformations, custom validation and anything reproducible all require code. Analytics roles are possible with heavy SQL and light Python; data science roles are not. ### How much Python do I need before starting a data science course? Enough to write a function, use a dictionary and a list comprehension, read an error message, and manipulate a DataFrame without following a tutorial line by line. Roughly eight weeks from zero. ### Should I learn object-oriented programming? Enough to read it, since libraries you use are built with it. You will write classes rarely in analysis work. Our [OOP guide](/articles/python-classes-objects) covers the level that is genuinely useful. ### Do AI coding assistants change how much I need to know? They lower the cost of writing code and raise the cost of not understanding it. Assistants can produce confidently wrong statistical code, such as a train-test split that leaks information, and you cannot catch that without the underlying knowledge. Use them for boilerplate and syntax; do not depend on them for statistical judgement. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/programming-for-data-science/). Cite with the title, "1stepGrow" and a link._ --- # Tech Career Roadmap: From Finished Course to First Offer > A tech career roadmap for the gap between finishing a course and an offer: public evidence, one target role, referrals, a tracked funnel and your first 90 days. - **Author:** Nishant Kiran — B2B SaaS & EdTech Content Strategist (https://www.1stepgrow.com/authors/nishant-kiran/) - **Published:** Aug 10, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Career Growth · **Format:** Roadmap · **Read time:** 6 min - **Canonical URL:** https://www.1stepgrow.com/articles/learning-to-earning-tech-career-roadmap/ ## Key takeaways - Most people over-invest in learning and under-invest in the six weeks that convert learning into an offer. - Applications are a numbers game only after positioning is right — a few dozen targeted applications usually beat hundreds of generic ones. - Referred applicants are more likely to be hired than similar non-referred applicants, so building a network is a stage, not an accident. - The first ninety days in the role shape your reputation, and your next raises, more than the offer negotiation did. This tech career roadmap covers the five to six months between finishing a course and accepting an offer. It turns knowledge into public evidence, evidence into conversations and conversations into interviews, at about ten to twelve hours a week, then covers your first ninety days in the job. Almost everyone who switches into tech hits the same moment. The course is finished, the certificate is downloaded, and then nothing happens, because nobody knows you exist. Capable people spend a year applying from here and get nowhere, usually because they improvise the conversion from learning to offer instead of planning it. This roadmap is for career switchers and new graduates who have the skills and not yet the job. It does not cover what to learn; the [roadmaps hub](https://www.1stepgrow.com/roadmaps) does that. It covers the six stages after learning, with overlapping timings, a checklist for portfolio evidence, and a table that diagnoses which part of your job search is failing. - **Total duration:** 5-6 months - **Weekly commitment:** 10-12 hours - **Typical entry salary:** Varies by track ### Stage 1: Close the evidence gap (Weeks 1-6) You have finished the course. Nobody can verify what you learned. This stage produces the artefacts that make your skill legible to a stranger reading for eight seconds. **Skills:** Project scoping, Technical writing, Documentation, Public deployment **Projects:** - One substantial project, deployed publicly, with a README explaining the trade-offs you chose - A written case study of that project — the problem, the approach, what broke, what you would change **Outcome:** A stranger can look at one link and understand what you can do without you being in the room. ### Stage 2: Position, then package (Weeks 7-9) Decide what you are applying for before you write anything. A CV aimed at everyone reaches no one. Pick one role title and one industry, and build the entire package around it. **Skills:** Role targeting, CV structure, LinkedIn positioning, Portfolio narrative **Projects:** - A one-page CV in reverse-chronological order, quantified, with no skills-bar graphics - A LinkedIn headline that names the role you want, not the one you are leaving **Outcome:** Your CV, LinkedIn and portfolio all tell the same story, and that story names a specific job. ### Stage 3: Build the surface area (Weeks 8-16, overlapping) Referred applicants are more likely to be hired than cold applicants. This stage runs in parallel with applications and is the one career switchers most consistently skip. **Skills:** Community participation, Writing in public, Informational interviews, Open source contribution **Projects:** - Two merged pull requests in a project you actually use - Four published posts about what you are building - Ten genuine conversations with people doing the job you want **Outcome:** People in your target field know your name before your application arrives. ### Stage 4: Apply like it is a system (Weeks 10-20) Volume without targeting is wasted. Targeting without volume is slow. Run both: a tracked pipeline, tailored applications, and a weekly review of what is converting. **Skills:** Application tracking, Cover letter writing, Company research, Follow-up discipline **Projects:** - A tracked pipeline of 25-40 targeted applications with response rates logged - One tailored paragraph per application referencing something specific about the company **Outcome:** You know your conversion rate at each stage, and you can see which stage is actually broken. ### Stage 5: Interview as a separate skill (Weeks 12-22, overlapping) Building and interviewing are different competencies. People who are good at the job routinely fail interviews for want of practice explaining themselves out loud. **Skills:** Technical interview practice, Behavioural stories (STAR), System design basics, Live coding under observation, Negotiation **Projects:** - Six mock interviews with someone who will be honest with you - Five prepared stories covering conflict, failure, ownership, ambiguity and impact **Outcome:** You can think out loud under pressure and explain a decision you made six months ago clearly. ### Stage 6: The first ninety days (Post-offer) The offer is the start of the compounding, not the end of the process. What you do in the first quarter sets your reputation and your trajectory for years. **Skills:** Onboarding fast, Asking good questions, Shipping early, Documenting as you learn **Projects:** - One small but visible improvement shipped in the first month - Written onboarding notes that help the next person **Outcome:** You are seen as someone who ships and improves things, which is what promotion cases are built from. ## How does the tech career roadmap fit into five to six months? The stages overlap on purpose, because networking and applying take time to produce results. Here is what a typical week focuses on across the plan: | Weeks | Main focus | Running in parallel | |---|---|---| | 1–6 | One deployed project and its case study | — | | 7–9 | Choose one role; rewrite CV and LinkedIn | Start community participation (week 8) | | 10–11 | First targeted applications | Networking, writing in public | | 12–16 | Applications and tracking | Networking; mock interviews begin | | 17–22 | Interviews and follow-ups | Applications continue to week 20 | | Post-offer | First ninety days | Onboarding notes, first shipped improvement | At ten to twelve hours a week, the 22 weeks before an offer add up to roughly 220–264 hours, which is less than most courses take, yet it is the part that decides whether the course pays off. ## Why does finishing a course not get you hired? Because nobody reading your application can see what you learned. That is the uncomfortable asymmetry at the heart of a career switch. You know exactly how much you learned: the concepts that clicked at 11pm, the bug that taught you more than the whole module, the moment the architecture finally made sense. A hiring manager scanning your CV for a few seconds knows none of that. They see a course name and a title, and they have seen hundreds of both this month. Everything in stage one exists to solve that asymmetry. A deployed project with a written explanation of your trade-offs converts private learning into public evidence. It is the single highest-leverage six weeks in this entire roadmap, yet it is the one people skip because it feels like it is not learning. ## What makes a portfolio project count as evidence? A portfolio project counts when a stranger can verify it without talking to you. Before you call stage one done, check that yours has all of these: - **A live link** that works on a phone, not only a repository. - **A README that opens with the problem**, then the approach and the trade-offs you chose. - **Something that went wrong**, and what you changed because of it. - **A realistic constraint**, such as real users, messy data or a hard performance limit. - **Readable history**: small commits with clear messages, so a reviewer can see how you work. ## Why do referrals matter so much? Referrals matter because they change who reads your application and how much they trust it. In a study of nine large firms across call centres, trucking and high-tech, [Burks, Cowgill, Hoffman and Housman (Quarterly Journal of Economics, 2015)](https://ideas.repec.org/a/oup/qjecon/v130y2015i2p805-839.html) found that referred applicants were more likely to be hired and more likely to accept offers than non-referred applicants, even though the two groups had similar skill characteristics. This is not because referral is unfair, exactly. A referral is a risk signal. Someone who works there is putting a small amount of their own reputation behind you, and that is worth more to a hiring manager than any certificate. Stage three is how you earn that, honestly and at scale: - **Contribute to projects people use.** GitHub's guide to [finding ways to contribute to open source](https://docs.github.com/en/get-started/exploring-projects-on-github/finding-ways-to-contribute-to-open-source-on-github) suggests searching for `good first issue` and `help wanted` labels. - **Write about what you are building**, including what broke. - **Have real conversations** with people doing the job. Not "can you refer me" on day one, but genuine participation over two months, after which asking is natural rather than awkward. ## How do you tell which part of your job search is broken? Track the conversion rate at each stage, not the number of applications sent. Once applications are running, most people track the wrong number. Total applications sent tells you nothing, whereas conversion at each stage tells you exactly what is broken. | What you see | What is usually broken | What to do | |---|---|---| | Almost no responses (roughly under one in twenty) | Positioning or CV | Rewrite around one named role; get cold reads | | Responses, but no technical rounds | Portfolio not carrying weight, or CV over-promising | Strengthen the headline project and its write-up | | Technical rounds, but no offers | Interview practice | Book mock interviews; rehearse explaining decisions | | Offers below expectation | Targeting and negotiation | Apply to a different band of company; prepare numbers | Each of those has a completely different fix, and none of them is "send more applications". Diagnose before you increase volume. ## How do employers screen career switchers? Most processes filter in a predictable order, so each stage of this roadmap maps onto one filter: 1. **Automated or recruiter CV screen.** The role title and relevant keywords must match, which is why stage two picks one role. 2. **Hiring manager review.** They open one link, so make sure it is your best project. 3. **Technical assessment.** A take-home task, live coding or a case, depending on the track. 4. **Behavioural interview.** Prepared STAR stories carry you here, especially about why you switched. 5. **References and offer.** People from stage three often become your most credible references. ## What can you earn once you land? What you can earn depends heavily on the track and the country, which is why the roadmap card says "varies by track". For scale, the latest US Bureau of Labor Statistics wage figures as of September 2026, which refer to May 2025, show how far apart tracks sit: | US occupation (BLS) | Median annual wage, May 2025 | |---|---| | [All computer and IT occupations](https://www.bls.gov/ooh/computer-and-information-technology/home.htm) | $109,470 | | All occupations | $50,980 | | [Software developers](https://www.bls.gov/ooh/computer-and-information-technology/software-developers.htm) | $135,980 | | [Data scientists](https://www.bls.gov/ooh/math/data-scientists.htm) | $120,230 | | [Web developers](https://www.bls.gov/ooh/computer-and-information-technology/web-developers.htm) | $92,650 | These are national medians for experienced and new workers combined, so a first role will usually start below them. Also, no equivalent official breakdown by tech role exists for India. ## Common mistakes between course and offer Across the stages, the same errors recur: - **Taking another course instead of starting stage one.** A second certificate does not fix an evidence gap. - **Applying to every title at once.** "Data analyst, data scientist or web developer" reads as no target at all. - **Networking only when you need something.** Relationships built during the job search work better when they started before it. - **Tracking volume instead of conversion.** It hides which stage is broken. - **Coasting after the offer.** The first ninety days set how your team sees you for years. ## Where to go next If you already have a finished project, start stage one by writing its case study this week. If you have not chosen a track yet, start with [Data Science vs Machine Learning vs AI](https://www.1stepgrow.com/articles/data-science-vs-machine-learning-vs-ai). For a technical path with stage gates, see the [full stack developer roadmap](https://www.1stepgrow.com/articles/full-stack-developer-roadmap-2026). And if you are switching later in your career, [How to switch to tech at 30](https://www.1stepgrow.com/articles/how-to-switch-to-tech-at-30) covers the specific constraints that come with that. ## Frequently asked questions ### How long does it take to get hired after finishing a course? For career switchers I have worked with, three to six months with a genuine portfolio was common, and longer without one. There is no reliable official statistic for this. The main variable is not skill level — it is whether you started building a network and public evidence during the course or waited until after it. ### How many applications should I send? Fewer than you think, better targeted than you think. Twenty-five to forty tailored applications with a tracked pipeline tend to outperform hundreds of generic ones. As a rule of thumb, if fewer than one in twenty applications gets any response, the problem is positioning, not volume, so fix the CV before sending more. ### Do cover letters still matter? A generic one does nothing. Two specific sentences about why this company and this role, linked to a relevant piece of your portfolio, can help at smaller companies where a person reads every application. At large employers with automated screening, the cover letter is rarely read, so spend the effort on the CV instead. ### Should I take a lower salary to break in? Sometimes, and it is a legitimate strategy — but cap it at one role and eighteen months, and negotiate the learning explicitly: the stack, the mentoring and the scope you will own. A first role that teaches you nothing is expensive at any salary, because your next offer is judged on what you did there. ### What if I get no responses at all? That is a positioning failure, not a skill failure, and it is fixable. Have three people in your target field read your CV cold and tell you what job they think you are applying for. If they cannot say, that is your problem. Rewrite the headline, summary and top project around one named role. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/learning-to-earning-tech-career-roadmap/). Cite with the title, "1stepGrow" and a link._ --- # AI in Healthcare: Where It Works, Where It Fails and Why > AI in healthcare, with the evidence: where it works, from retinal screening to AI scribes, why sepsis and bias failures happened, and what the field demands. - **Author:** Althaf Ashraf — AI Systems Engineer, Tata Consultancy Services (https://www.1stepgrow.com/authors/althaf-ashraf/) - **Published:** Aug 9, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Artificial Intelligence · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/ai-in-healthcare/ ## Key takeaways - The wins are concentrated in narrow, well-specified tasks with clear ground truth — imaging, triage, documentation. - The FDA said in January 2025 that it had authorised more than 1,000 AI-enabled devices, and radiology dominates its public list. - Most failures are deployment failures, not modelling failures: distribution shift, workflow mismatch and no clinician buy-in. - Domain knowledge is worth more in healthcare AI than in almost any other vertical. - Regulatory and validation literacy is a genuine differentiator for anyone entering this field. AI in healthcare works best on narrow, well-defined tasks with clear ground truth, such as screening retinal images or drafting clinical notes. It fails most often after deployment, not in testing: when patients differ from the training data, when tools do not fit clinical workflow, or when a convenient proxy hides bias. The gap between the two is rarely the model. When one US health system independently validated Epic's proprietary sepsis model, it missed 67% of patients with sepsis, well short of the developer's reported performance. What separated failures like that from the successes was everything that happened after the paper. This guide is for engineers, analysts and clinicians weighing work in the field. It sets out the documented wins and the documented failures, each with its source, and then the skills that decide which side of that line a project lands on. ## Where does AI in healthcare genuinely work? The successes share a pattern: a narrow task, clear ground truth and a human in the loop. ### Imaging and regulated devices **Medical imaging.** This is the clearest success. In 2018 the FDA authorised IDx-DR, which screens retinal photographs for diabetic retinopathy. Its [pivotal trial](https://pmc.ncbi.nlm.nih.gov/articles/PMC6550188/) enrolled 900 participants across 10 primary care sites. It reported 87.2% sensitivity and 90.7% specificity. The authors also describe it as the first autonomous diagnostic AI system authorised by the FDA in any field of medicine. Flagging findings on radiographs and screening pathology slides follow the same template: a well-specified task, abundant labelled data and unambiguous ground truth. **Regulated devices at scale.** Imaging is no longer a handful of pilots. By January 2025 the FDA said it had [authorised more than 1,000 AI-enabled devices](https://www.fda.gov/news-events/press-announcements/fda-issues-comprehensive-draft-guidance-developers-artificial-intelligence-enabled-medical-devices), and radiology accounts for most entries on its public [list of AI-enabled medical devices](https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices). ### Prediction, documentation and admin **Deterioration prediction.** Models that watch vitals and lab results can flag patients at risk of sepsis or decline. They add value where they are integrated properly. They also show why alert design matters: a model that fires too many alerts gets ignored, which can make it worse than nothing. **Clinical documentation.** Ambient AI scribes draft notes from a recorded consultation, and the clinician reviews them before they count. The evidence is promising but more modest than vendor claims. In a [randomised trial of 238 outpatient physicians](https://pmc.ncbi.nlm.nih.gov/articles/PMC12768499/) published in NEJM AI, one scribe (Nabla) cut time spent writing notes by 9.5% compared with usual care, while the other (Microsoft DAX Copilot) showed no significant change. Both tools showed possible improvements in burnout measures that still need confirming in larger trials. Clinicians also noted occasional clinically significant inaccuracies, so the review step is not optional. **Administrative work.** Coding, scheduling, prior authorisation and claims are unglamorous, but the return on investment is easy to measure and the clinical risk is comparatively low. | Use case | Real example | What the evidence shows | Main risk | |---|---|---|---| | Retinal screening | IDx-DR (2018) | 87.2% sensitivity, 90.7% specificity in a pivotal trial | Ungradable images, access to follow-up care | | Radiology support | Devices on the FDA list | Largest category of FDA-authorised AI devices | Performance drift across scanners and sites | | Sepsis prediction | Epic Sepsis Model | Poor discrimination in external validation | Missed cases and alert fatigue | | Ambient documentation | DAX Copilot, Nabla | 9.5% less note time for one tool, no significant change for the other | Occasional inaccuracies in drafted notes | ## Why do healthcare AI projects keep failing? Most failures are deployment failures. The model looked good on paper, but the real world did not match the paper. ### When the data or the setting changes **Distribution shift.** Distribution shift means the data a model meets in use differs from the data it learned from, and it is one of the most common failure modes. A model trained at one hospital meets a different population, different scanners and different labelling conventions. It then degrades quietly. A 2018 [PLOS Medicine study of pneumonia detection on chest X-rays](https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1002683) found that models performed better internally than at external hospitals in 3 of 5 comparisons. The network could also identify which hospital system an image came from with near-perfect accuracy. It was learning the site, not just the disease. **Validation that does not survive contact.** The best-known sepsis example is Epic's proprietary sepsis model. An [external validation in JAMA Internal Medicine](https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2781307) covering 38,455 hospitalisations at Michigan Medicine found an area under the curve of 0.63, well below the developer's reported figures. The model missed 67% of patients with sepsis, yet it still generated alerts in 18% of all hospitalisations. ### When people and proxies are ignored **Workflow mismatch.** A tool that is accurate but adds three clicks to an already compressed consultation will not be used. Clinical workflow is not a deployment detail; it decides whether the model has any effect at all. **Label quality and proxy bias.** Healthcare labels are often proxies. "Diagnosed with X" means someone was coded with X, which reflects who got tested, who had access, and how the coding was done. The landmark example is a widely used care-management algorithm studied in [Science in 2019](https://pubmed.ncbi.nlm.nih.gov/31649194/). It predicted healthcare costs rather than illness, and because less money was spent on Black patients with the same needs, it under-referred them. Fixing the bias would have raised the share of Black patients receiving additional help from 17.7% to 46.5%. **No clinician ownership.** Projects built by data teams and presented to clinicians at the end have a poor record. The ones that work involve clinicians in choosing the problem, not just validating the answer. **Equity failures.** Models trained on populations that under-represent certain groups can perform worse for those groups. In healthcare that is not a metrics footnote. It is a harm, and increasingly a regulatory matter too. ## What does working in healthcare AI require? Healthcare AI rewards domain fluency, validation literacy and patience more than raw modelling skill. Four things matter more here than in most other areas of applied AI. - **Domain fluency.** Understanding why a lab value is missing — often the test was not ordered, which itself carries information — is the kind of knowledge that separates a useful model from a dangerous one. It takes deliberate effort, but it compounds. - **Validation literacy.** Know why a strong test-set score is not clinical validation, what external and prospective validation involve, and when a tool becomes a regulated device. The sepsis and pneumonia studies above are the reason this matters, and understanding them puts you ahead of most candidates entering this space. - **Patience with long cycles.** Procurement is slow, validation is slow, and integration with hospital systems is slow. If you need a fast feedback loop, this vertical will frustrate you. - **A long-term view.** The trade-off is durability. Generic ML skill is increasingly commoditised, whereas clinical domain expertise combined with ML is not, and it is hard to displace. Who should not choose this field? Anyone who wants to ship weekly, or who is uncomfortable being accountable for errors that affect patients rather than dashboards. ## Is AI in healthcare living up to the promise? Partly. AI in healthcare is neither the revolution the 2018 headlines promised nor the disappointment the backlash suggested. As of September 2026, it is a field where narrow, carefully deployed tools deliver real value, while ambitious general-purpose systems keep meeting the same wall. For anyone entering, the lesson is consistent. Pick a narrow problem, work with the clinicians who have it, and treat deployment as the hard part rather than the afterthought. That is where the successes have consistently come from. ## Where to go next To test your interest, read the Epic sepsis validation linked above in full; it is the clearest short lesson in how clinical AI goes wrong. For the technical path, see our [AI engineer roadmap](https://www.1stepgrow.com/articles/ai-engineer-roadmap-2026). For how the field's terms relate, read [AI vs machine learning vs deep learning](https://www.1stepgrow.com/articles/ai-vs-machine-learning-vs-deep-learning). For the forecasting side of clinical data, [predictive analytics in healthcare](https://www.1stepgrow.com/articles/predictive-analytics-in-healthcare) goes deeper, and [what data science is and why it matters](https://www.1stepgrow.com/articles/what-is-data-science) covers the underlying discipline. ## Frequently asked questions ### What is AI actually good at in healthcare today? Narrow, well-defined tasks with reliable ground truth: detecting specific findings in medical images, flagging deterioration risk from vitals, drafting clinical documentation for a clinician to review, and administrative work like coding and scheduling. Broad diagnostic reasoning remains far harder than headlines suggest, and even documentation tools show modest, uneven gains in trials. ### Why do so many healthcare AI projects fail? Rarely because the model was inaccurate in testing. Usually the deployment population differed from the training population, the tool did not fit clinical workflow, or clinicians were brought in too late to trust or use it. Epic's sepsis model is the standard example: external validation found it missed most sepsis cases while still alerting on a large share of admissions. ### Do I need a medical background to work in healthcare AI? No, but you need to build domain fluency deliberately and work closely with clinicians. Teams that treat healthcare as just another tabular dataset produce models that are technically sound and clinically useless. A practical start is learning why values go missing, how diagnoses get coded, and which decision a clinician actually makes at the point your model would fire. ### Is healthcare AI a good career area? Yes, if you can accept slow cycles. Procurement, validation and hospital integration all take time, and regulatory overhead is real. In exchange, clinical domain expertise compounds unusually well and is much harder to displace than generic ML skill. It suits people who prefer durable, accountable work over shipping something new every week. ### What regulatory knowledge is needed? Enough to know when a tool becomes a regulated medical device, what clinical validation requires beyond a test-set score, and how data governance and patient privacy constrain what you can build. In practice that means understanding external and prospective validation, not just held-out accuracy. Few candidates arrive with this literacy, so it is a strong differentiator. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/ai-in-healthcare/). Cite with the title, "1stepGrow" and a link._ --- # The Future of Programming Languages: What Is Actually Changing > The future of programming languages, using TIOBE, Stack Overflow and GitHub data: why Python and TypeScript lead, types are winning, and what to learn. - **Author:** Sivaranjani S — Cloud Operations Engineer, Zoho (https://www.1stepgrow.com/authors/sivaranjani-s/) - **Published:** Aug 8, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/future-of-programming-languages/ ## Key takeaways - Language choice is converging on ecosystem strength, not language features — which is why Python keeps winning despite its flaws. - As of September 2026, Python ranks first on the TIOBE Index, while GitHub's Octoverse 2025 found TypeScript overtook Python by contributor count in August 2025. - In the 2025 Stack Overflow Developer Survey, 84% of respondents used or planned to use AI tools, yet 46% distrusted their accuracy, so reviewing code matters more. - Type systems are winning across the board: TypeScript, typed Python, Rust, and gradual typing everywhere else. - Learn one language deeply before learning three shallowly — the concepts transfer, the fluency does not. The future of programming languages looks less like a revolution and more like consolidation. Python leads for data and AI, TypeScript leads for products, typed languages keep gaining ground, and AI assistants reward conventional, readable code. For most learners, the practical answer is to pick one language by destination and learn it deeply. The catch is that the popular rankings disagree. TypeScript sits outside the TIOBE top 20, yet it became the most used language on GitHub in 2025. Which chart you read can change what you decide to learn. This guide is for students, career changers and working developers choosing their next language. It reads TIOBE, Stack Overflow and GitHub data side by side, separates what AI assistants really change from the hype, and ends with five rules for choosing, plus who should ignore them. ## Which programming languages lead in 2026? As of September 2026, Python ranks first on TIOBE, JavaScript is the most used language among Stack Overflow's 2025 respondents, and TypeScript leads GitHub by contributors. The three sources measure different things, so they disagree in useful ways. The [TIOBE Index](https://www.tiobe.com/tiobe-index/) tracks search-engine popularity, the [Stack Overflow Developer Survey](https://survey.stackoverflow.co/2025/technology) asks developers what they actually used, and GitHub's Octoverse counts contributors on its platform. | Language | TIOBE rank, September 2026 | Used by (Stack Overflow 2025) | Best suited to | |---|---|---|---| | Python | 1 | 57.9% | Data, ML, AI, scripting, backend | | JavaScript | 6 | 66.0% | Anything that runs in a browser | | SQL | 8 | 58.6% | Querying data, in every track | | TypeScript | Outside top 20 | 43.6% | Web products, frontend and backend | | Java | 4 | 29.4% | Enterprise systems, Android | | C# | 5 | 27.8% | Enterprise, Windows, game development | | Go | 12 | 16.4% | Backend services, cloud tooling | | Rust | 10 | 14.8% | Systems, infrastructure, performance-critical code | Read the columns together rather than separately. For instance, TypeScript barely registers on TIOBE's search-based measure, yet GitHub's [Octoverse 2025 report](https://github.blog/news-insights/octoverse/octoverse-a-new-developer-joins-github-every-second-as-ai-leads-typescript-to-1/) found that in August 2025 it overtook Python as the most used language on GitHub by contributor count. Similarly, Rust climbed from 18th to 10th on TIOBE over the past year, while Go slipped from 8th to 12th. ## Why do ecosystems beat languages? Languages win on libraries, tooling and hiring pools, not on elegance. That has been the most reliable pattern of the last decade. Python is the clearest case. It is slow, its packaging has been a running joke for years, and its concurrency story is awkward. Yet it dominates data, machine learning and AI because that is where the libraries are, and new AI platforms typically ship a Python SDK early, which reinforces the position further. The 2025 Stack Overflow survey recorded a 7 percentage point rise in Python use from 2024 to 2025, which it linked to AI, data science and back-end work. The lesson for language choice is unromantic. Ask where the libraries for your problem live, and go there. Language aesthetics are a rounding error next to ecosystem depth. ## Why are typed languages winning? Types catch a whole category of error before code ships, and they act as documentation that cannot go stale. Both matter more as codebases, teams and the share of AI-generated code grow. A decade ago dynamic typing was ascendant; that has reversed. TypeScript is now the professional default for anything JavaScript-adjacent. Python added optional type hints through [PEP 484](https://peps.python.org/pep-0484/), and serious codebases use them. Rust, meanwhile, made an aggressive type system a selling point rather than an obstacle — and developers like it, since Stack Overflow's 2025 survey again named Rust the most admired language, at 72%. The reason is scale, both of codebases and of teams. Types are documentation that cannot go stale, and they catch a category of error before it ships. The AI angle reinforces this. GitHub's Octoverse report describes TypeScript's rise as part of a shift towards typed languages that make agent-assisted coding more reliable in production. When a meaningful share of your code is generated, a type checker is one of the cheapest ways to catch the confident, plausible mistakes that assistants produce. ## What does AI actually change about programming? AI makes writing conventional code much cheaper and makes reviewing it more valuable. Most commentary gets this wrong in one direction or the other. **What changed:** the cost of writing conventional code fell dramatically. Boilerplate, transformations, glue code, translating between formats and remembering plotting syntax are all much faster. Adoption is broad, too: 84% of respondents to the [2025 Stack Overflow survey](https://survey.stackoverflow.co/2025/ai) were using or planning to use AI tools in development. **What did not change:** deciding what to build, designing how the pieces fit, knowing whether the result is correct, and debugging something subtly wrong at 3am. The second-order effect is the interesting one. If writing code is cheap and reviewing it is not, then the bottleneck moves to review. The same survey found that 46% of developers distrusted the accuracy of AI tools, compared with 33% who trusted it. Reviewing well requires you to understand the code thoroughly enough to spot what an assistant got confidently wrong. As a result, the floor rose. Trivial CRUD work is genuinely being automated, while understanding systems is worth more, not less. There is also a quiet consequence for language choice. Assistants learn from public code, so it is reasonable to expect them to be strongest in languages with large public corpora and strong conventions — Python, TypeScript, Java and Go — and weaker in niche languages. ## What does the future of programming languages mean for each major language? **Python** stays dominant in data, ML and AI, and it remains a strong general-purpose choice. It is not going anywhere. **TypeScript** owns frontend and a large share of backend. It is the safest choice for anyone building products. **Rust** keeps growing in infrastructure, systems and performance-critical work, yet it remains a poor first language. **Go** remains a pragmatic choice for backend services and cloud tooling — boring in the way infrastructure should be — even though its search popularity has dipped. **Java and C#** are not going anywhere, whatever the internet says. Both sit in TIOBE's top five, power enormous enterprise codebases, and have modernised considerably. **SQL** is the language most people forget to count, although most data professionals use it more than anything else. It rose to 8th on TIOBE in September 2026. ## What should you actually learn? Start with these five rules. 1. **Learn one language deeply.** Deep knowledge of one transfers as concepts — scope, memory, concurrency, abstraction — and those concepts make the second language take weeks rather than months. By contrast, three shallow languages transfer as syntax you could have looked up. 2. **Pick it by destination.** Data, ML or AI: Python. Web products: TypeScript. Systems and infrastructure: Go, then Rust. Enterprise: Java or C#. 3. **Learn SQL regardless.** Whatever your track, you will use it. 4. **Get comfortable with types.** Whichever language you choose, use its type system, because that is where the industry has landed. 5. **Use the assistants, and read what they write.** The skill that is appreciating is review, and you cannot review what you do not understand. Who should break these rules? Embedded and games developers, mainly. If that is your destination, then C, C++ or C# is the right first language, whatever the general rankings say. ## Does the language you pick matter that much? Less than language wars suggest. The engineers whose careers compound are not the ones who picked the winning syntax. Rather, they are the ones who understood systems, data and trade-offs well enough that the language became an implementation detail. Pick the one with the ecosystem for your problem, learn it properly, and spend the energy you saved on the parts that transfer. ## Related reading For where Python sits in data work specifically, see [the role of programming in data science](https://www.1stepgrow.com/articles/programming-for-data-science). To start with Python properly, our [building blocks guide](https://www.1stepgrow.com/articles/python-basics) is the entry point, and the [full stack developer roadmap](https://www.1stepgrow.com/articles/full-stack-developer-roadmap-2026) shows where TypeScript fits in web work. ## Frequently asked questions ### Will AI replace programmers? It has replaced a share of the typing, not the deciding. What is being automated is boilerplate and translation between well-understood forms. What is not is deciding what to build, how the pieces fit and whether the result is correct — and reviewing AI output requires more judgement, not less. ### Which programming language should I learn first in 2026? Python, for most people. It has the largest job market across data, AI, backend and scripting, gentle syntax and the best learning resources, and it tops the TIOBE Index in September 2026. Learn it properly before adding a second. The exception is web products, where TypeScript is the better first choice. ### Is Rust worth learning? If you work on systems, infrastructure or performance-critical code, increasingly yes. For application and data work it is a poor first choice — the borrow checker is a steep tax to pay for a problem you may not have. ### Is JavaScript still necessary? For anything running in a browser, yes, and that has not changed. TypeScript has effectively become the professional default, and learning it means learning JavaScript with guardrails. JavaScript was also the most used language in the 2025 Stack Overflow Developer Survey, at 66% of respondents, so it is not fading. ### How many languages should a developer know? One deeply, one or two adequately. Depth transfers as concepts; breadth mostly transfers as syntax you can look up. Interviewers test the language you claim, so claim carefully. SQL is the sensible addition for almost everyone, because it is used across data, backend and analytics roles whatever your main language is. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/future-of-programming-languages/). Cite with the title, "1stepGrow" and a link._ --- # NumPy Tutorial for Beginners: NumPy Arrays in Python (Part 1) > NumPy tutorial for beginners: what an array is, why it ran about 20 times faster than a list here, how one stray string turns numbers into text, and more. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Aug 7, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/numpy-tutorial/ ## Key takeaways - A NumPy array is a fixed-type, contiguous block of memory, and that layout is the main reason it is fast. - Every array has one dtype. Mixing types silently upcasts, so a single stray string turns a numeric array into text. - Vectorised operations replace loops; if you are writing a for loop over an array, there is usually a shorter and faster way. - Shape is the concept everything else in NumPy builds on, so check it first when an error makes no sense. This NumPy tutorial starts with the array: a fixed-type, contiguous block of memory that makes numerical Python fast. Doubling a million integers ran roughly 20 times faster as an array than as a list in the timing below. pandas, scikit-learn, PyTorch and matplotlib all build on it. You can use those libraries without knowing NumPy, right up until an error mentions a shape or a dtype. One stray string in a column of numbers, for instance, silently turns the whole array into text. Part one is for Python beginners moving into data work: what arrays are, why they are fast, how to create them and how to avoid the dtype trap. [Part two](https://www.1stepgrow.com/articles/numpy-indexing-slicing) moves on to indexing and slicing. If you have not set up Python yet, follow the [Anaconda on Windows guide](https://www.1stepgrow.com/articles/install-anaconda-windows) or the [macOS version](https://www.1stepgrow.com/articles/install-anaconda-mac) first. As of September 2026 the latest release is NumPy 2.5. The outputs in this NumPy tutorial come from NumPy 2.2 and match the behaviour described in the current [NumPy 2.x documentation](https://numpy.org/doc/stable/). ## Why is a NumPy array faster than a Python list? A Python list is a sequence of pointers. Each element is a full Python object stored somewhere in memory, and the list holds their addresses. By contrast, a NumPy array is a single contiguous block of memory holding raw values of one fixed type. That difference has two consequences. The array uses far less memory, and operations on it run as compiled loops over adjacent memory rather than as interpreted Python. ```python import numpy as np import time py_list = list(range(1_000_000)) np_array = np.arange(1_000_000) start = time.perf_counter() result = [x * 2 for x in py_list] print(f"list: {time.perf_counter() - start:.4f}s") start = time.perf_counter() result = np_array * 2 print(f"array: {time.perf_counter() - start:.4f}s") ``` Timed properly with `timeit` on NumPy 2.2, the array version was roughly 20 times faster on a million integers. The exact ratio depends on your hardware and the operation, but it grows with the size of the data. ### Worked example: the memory difference ```python import sys print(np_array.nbytes) # 8000000 - 8 bytes per int64 print(sys.getsizeof(py_list)) # 8000056 - the pointers alone print(sys.getsizeof(1)) # 28 - each int object on top of that ``` The list's 8 MB covers only the pointers. Each integer it points to is a separate object of about 28 bytes on CPython, so the real total is several times the array's 8 MB. ## How do you create NumPy arrays? From an existing sequence: ```python import numpy as np a = np.array([1, 2, 3, 4]) print(a) # [1 2 3 4] print(a.dtype) # int64 b = np.array([[1, 2, 3], [4, 5, 6]]) print(b.shape) # (2, 3) print(b.ndim) # 2 print(b.size) # 6 ``` The default integer is `int64` on every 64-bit platform, including Windows, since [NumPy 2.0](https://numpy.org/doc/stable/release/2.0.0-notes.html). Older tutorials that show `int32` on Windows describe NumPy 1.x. From scratch, which is what you will use most: ```python np.zeros((2, 3)) # 2x3 of 0.0 np.ones(5) # [1. 1. 1. 1. 1.] np.full((2, 2), 7) # 2x2 of 7 np.eye(3) # 3x3 identity np.empty(3) # uninitialised - contains whatever was in memory np.arange(0, 10, 2) # [0 2 4 6 8] - like range() np.linspace(0, 1, 5) # [0. 0.25 0.5 0.75 1. ] - n evenly spaced points ``` `arange` takes a step, whereas `linspace` takes a count. The [`arange` documentation](https://numpy.org/doc/stable/reference/generated/numpy.arange.html) itself recommends `linspace` for non-integer steps, because floating-point steps can overshoot: ```python np.arange(1, 1.3, 0.1) # [1. 1.1 1.2 1.3] - four values, the stop leaked in np.linspace(1, 1.2, 3) # [1. 1.1 1.2] - exactly three, as asked ``` For random data, use the generator API: ```python rng = np.random.default_rng(seed=42) rng.random(3) # [0.77395605 0.43887844 0.85859792] rng.integers(0, 10, size=5) # [0 6 2 0 5] rng.normal(loc=0, scale=1, size=4) # four draws from a standard normal ``` Always seed when you want reproducible results. The older `np.random.rand` style still works, but [`default_rng` is the recommended constructor](https://numpy.org/doc/stable/reference/random/generator.html) in current NumPy. ## What is the dtype upcasting trap? Every array has exactly one dtype. Put mixed types in and NumPy silently promotes to whatever holds them all: ```python np.array([1, 2, 3]).dtype # int64 np.array([1, 2, 3.0]).dtype # float64 - one float promotes everything np.array([1, 'a']).dtype # 2 # [False False True True] b = np.array([10, 20, 30, 40]) a + b # [11 22 33 44] - elementwise a * b # [ 10 40 90 160] ``` Note that `*` is elementwise multiplication, not matrix multiplication. For the matrix product, use `@` or `np.matmul`, which [Part 4 of this series](https://www.1stepgrow.com/articles/numpy-aggregation-functions) covers alongside the rest of the linear algebra. The rule of thumb: **if you are writing a Python `for` loop over a NumPy array, there is almost always a vectorised way that is both faster and shorter.** ```python arr = np.array([1, 2, 3]) # don't result = [] for x in arr: result.append(x * 2 + 1) # do result = arr * 2 + 1 # [3 5 7] ``` ## How do you check an array's shape and dtype? Read the `.shape`, `.ndim`, `.size` and `.dtype` attributes. Those four answer most questions: ```python arr = np.arange(12).reshape(3, 4) arr.shape # (3, 4) - length along each axis arr.ndim # 2 - number of axes arr.size # 12 - total elements arr.dtype # int64 - element type ``` `shape` is the one you will check constantly. Most NumPy errors are shape mismatches, so reading the shape first turns a confusing traceback into an obvious problem. One NumPy 2 detail surprises people in notebooks. If a cell ends with a single element, such as `arr[0, 0]`, Jupyter shows `np.int64(0)` rather than `0`. That is the new scalar representation, not a bug; `print()` still shows the plain number. ## NumPy array quick reference | Task | Function | Example result | |---|---|---| | Array from a list | `np.array([1, 2, 3])` | `[1 2 3]` | | Filled array | `np.zeros`, `np.ones`, `np.full` | `np.full((2, 2), 7)` gives a 2x2 of 7 | | Integer range | `np.arange(0, 10, 2)` | `[0 2 4 6 8]` | | N evenly spaced points | `np.linspace(0, 1, 5)` | `[0. 0.25 0.5 0.75 1.]` | | Reproducible random numbers | `np.random.default_rng(42)` | a seeded `Generator` | | Change type | `arr.astype(np.float64)` | a new array | | Inspect | `.shape`, `.ndim`, `.size`, `.dtype` | `(3, 4)`, `2`, `12`, `int64` | ## Common mistakes with NumPy arrays Most beginners hit the same handful of problems, and each one is covered earlier in this NumPy tutorial: - **Using `np.empty` as if it were `np.zeros`.** It skips initialisation, so the values are whatever was in memory. - **Using `arange` with a float step.** Use `linspace` instead when you need an exact number of points. - **Loading mixed data into one array.** One text value turns everything into strings, so clean the data first. - **Expecting `astype(int)` to round.** It truncates; call `np.round` first if you want rounding. - **Writing loops over arrays.** Vectorise instead, because the loop throws away the speed you came for. ## What comes next in this NumPy tutorial? [Part two](https://www.1stepgrow.com/articles/numpy-indexing-slicing) covers indexing and slicing, including the critical detail that NumPy slices are *views* into the original array, not copies. That behaviour surprises nearly everyone the first time they modify one. ## Related reading [NumPy Part 2: array indexing and slicing](https://www.1stepgrow.com/articles/numpy-indexing-slicing) continues the series, then [Part 3 on broadcasting](https://www.1stepgrow.com/articles/numpy-broadcasting). For how Python's own number types behave, see [Python numeric data types](https://www.1stepgrow.com/articles/python-numeric-data-types). For where NumPy sits in the wider stack, see [tools and techniques for data science](https://www.1stepgrow.com/tools-and-techniques-for-data-science), and for a one-page summary the [NumPy cheat sheet](https://www.1stepgrow.com/articles/numpy-cheat-sheet) is the companion reference. ## Frequently asked questions ### Why use NumPy instead of Python lists? There are two reasons. First, memory: an array stores raw values contiguously, whereas a list stores pointers to separate Python objects. Second, speed: array operations run as compiled loops over the whole block instead of interpreting Python bytecode per element. The gap is usually between ten and a hundred times, depending on the operation and the hardware. ### What is a dtype in NumPy? A dtype is the single data type shared by every element in an array, such as int64, float64 or bool. It is fixed when the array is created, which is what makes the memory layout predictable and the operations fast. If you need a different type, astype() creates a new array rather than changing the existing one in place. ### Do I need NumPy if I already use pandas? pandas is built on NumPy, so you are already using it whenever you use a DataFrame. You need enough NumPy to understand dtypes, shapes and broadcasting, because pandas error messages and performance advice refer to those concepts constantly. Learning the array basics first makes pandas behaviour far less mysterious. ### What is the difference between an array's size and its shape? Shape is a tuple giving the length along each dimension, for example (3, 4) for three rows and four columns. Size is the total number of elements, which is the product of the shape, so 12 in that example. The ndim attribute tells you how many dimensions there are, which is simply the length of the shape tuple. ### Why does NumPy print np.int64(10) instead of 10? Since NumPy 2.0, the repr of a NumPy scalar includes its type, so a notebook cell ending in a[0] shows np.int64(10). The value is still ten. print() and str() show the plain number, and you can convert explicitly with int() or float() when you need a built-in Python type. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/numpy-tutorial/). Cite with the title, "1stepGrow" and a link._ --- # NumPy Indexing and Slicing: Views, Masks and Fancy Indexing (Part 2) > NumPy indexing and slicing explained: why editing a slice changes your original array, when masks and fancy indexing copy instead, and how to avoid both bugs. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Aug 6, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/numpy-indexing-slicing/ ## Key takeaways - Basic slicing returns a view that shares memory with the original, so writing to it changes the source array. - Boolean masking and fancy indexing return copies, not views. - Boolean masks are the idiomatic way to filter, and they combine with & and | rather than with and/or. - When in doubt, call .copy(), because a defensive copy is cheaper than a bug you cannot reproduce. NumPy indexing and slicing pulls data out of arrays with `b[row, col]`, ranges like `a[1:4]`, boolean masks like `a[a > 25]`, and lists of positions. It looks like list indexing, with one big difference: a slice is a view onto the same memory, so writing to it changes the original array. That rule catches people who clean a slice of a dataset and later find the raw data altered. Masks and fancy indexing do the opposite and return copies. This part of the series is for readers who can create arrays, as covered in [Part 1](https://www.1stepgrow.com/articles/numpy-tutorial). It shows each kind of indexing with real output, how to tell a view from a copy, and the mistakes behind most indexing bugs. Every output below comes from running the code on NumPy 2.2. The rules themselves are documented in the official [NumPy indexing guide](https://numpy.org/doc/stable/user/basics.indexing.html). ## How does basic indexing work in NumPy? One dimension behaves like a list: ```python import numpy as np a = np.array([10, 20, 30, 40, 50]) a[0] # 10 a[-1] # 50 a[1:4] # [20 30 40] a[::2] # [10 30 50] a[::-1] # [50 40 30 20 10] ``` Multiple dimensions, however, use a comma rather than chained brackets: ```python b = np.arange(12).reshape(3, 4) # [[ 0 1 2 3] # [ 4 5 6 7] # [ 8 9 10 11]] b[1, 2] # 6 - row 1, column 2 b[1] # [4 5 6 7] - whole row b[:, 1] # [1 5 9] - whole column b[0:2, 1:3] # [[1 2] # [5 6]] ``` `b[1][2]` also works, but it first creates a temporary array for `b[1]` and then indexes that. As a result, `b[1, 2]` is the idiomatic and faster form. In a notebook, a single element displays as `np.int64(6)` rather than `6`. That is the NumPy 2 scalar representation; the value is the same. ## Why does changing a slice change the original array? Because basic slicing returns a view that shares memory with the source array, not a copy. This is the single most important thing in this article. ```python original = np.array([1, 2, 3, 4, 5]) piece = original[1:4] piece[0] = 999 print(original) # [ 1 999 3 4 5] <- the source changed ``` Basic slicing returns a **view**: a window onto the same memory. Nothing is copied, which is why slicing a million-element array is instant. That is a feature, until you slice a dataset, clean the slice, and then discover you have modified the raw data you meant to preserve. Check with `.base`: ```python piece.base is original # True - it is a view original.base is None # True - it owns its data ``` Take an explicit copy when you need independence: ```python original = np.array([1, 2, 3, 4, 5]) safe = original[1:4].copy() safe[0] = 999 print(original) # [1 2 3 4 5] - untouched this time ``` The rule: **if you are going to write to it, copy it.** The [copies and views page](https://numpy.org/doc/stable/user/basics.copies.html) in the NumPy docs lists which other operations, such as `reshape` and `ravel`, return views where possible. ## How do you filter a NumPy array with a condition? Use boolean masking, the idiomatic way to filter. A comparison produces a boolean array, and indexing with that array selects the positions where it is `True`. ```python temps = np.array([18, 25, 31, 12, 28, 35]) mask = temps > 25 print(mask) # [False False True False True True] print(temps[mask]) # [31 28 35] # usually written inline print(temps[temps > 25]) ``` Combine conditions with `&`, `|` and `~`, **not** `and`, `or` and `not`: ```python temps[(temps > 20) & (temps < 32)] # [25 31 28] temps[(temps < 15) | (temps > 30)] # [31 12 35] temps[~(temps > 25)] # [18 25 12] ``` Notice that the result keeps the original order: `31` comes before `12` because it appears earlier in `temps`. The parentheses are required. Because `&` binds more tightly than `>`, the expression `temps > 20 & temps < 32` parses as something quite different and raises an error. Python's `and` fails here for a related reason. It needs a single boolean from an array of many, which NumPy refuses to guess, hence the familiar "truth value of an array with more than one element is ambiguous". Masks also work on the left of an assignment: ```python scores = np.array([45, 82, 91, 38, 67]) scores[scores < 50] = 0 print(scores) # [ 0 82 91 0 67] ``` Meanwhile, `np.where` gives a vectorised if/else: ```python np.where(scores >= 50, 'pass', 'fail') # ['fail' 'pass' 'pass' 'fail' 'pass'] ``` ## What is fancy indexing in NumPy? Fancy indexing means indexing with a list or array of positions. It selects elements in any order, with repeats, and always returns a copy: ```python a = np.array([10, 20, 30, 40, 50]) a[[0, 3, 4]] # [10 40 50] a[[2, 2, 0]] # [30 30 10] - repeats allowed ``` In two dimensions, paired index arrays select individual elements: ```python b = np.arange(12).reshape(3, 4) rows = [0, 1, 2] cols = [1, 2, 3] b[rows, cols] # [ 1 6 11] - (0,1), (1,2), (2,3) ``` Fancy indexing and boolean masking both return **copies**, not views. The selected elements are not generally evenly spaced, so there is nothing to view. That is the reverse of the slicing rule, and it is worth remembering. ```python f = a[[0, 1]] np.shares_memory(f, a) # False - a copy np.shares_memory(a[1:3], a) # True - a view ``` ## How do you change part of a NumPy array in place? Assign to a slice. The value on the right is broadcast across the selected positions: ```python b = np.zeros((3, 4)) b[1, :] = 5 # whole row becomes 5 b[:, 0] = [1, 2, 3] # column gets these values b[0:2, 0:2] = 9 # block becomes 9 ``` This is efficient, since it writes in place rather than building a new array. [Part 3](https://www.1stepgrow.com/articles/numpy-broadcasting) explains the broadcasting rules that decide which shapes are allowed on the right-hand side. ## Worked example: filtering rows of a table Filtering rows of a 2D array by a condition on one column: ```python data = np.array([ [1, 25, 50000], [2, 32, 72000], [3, 19, 31000], [4, 45, 95000], ]) # rows where age (column 1) is over 30 adults = data[data[:, 1] > 30] # [[ 2 32 72000] # [ 4 45 95000]] ``` `data[:, 1] > 30` produces a mask of length 4, one value per row. Using it as the row index then selects the matching rows whole. Because this is boolean masking, `adults` is a copy, so you can edit it without touching `data`. ## NumPy indexing and slicing quick reference | Syntax | Selects | Returns | |---|---|---| | `a[2]`, `b[1, 2]` | one element | a scalar | | `a[1:4]`, `b[:, 1]` | a range or row/column | a view | | `a[::-1]` | reversed range | a view | | `a[a > 25]` | elements matching a mask | a copy | | `a[[0, 3, 4]]` | listed positions | a copy | | `np.where(cond, x, y)` | elementwise if/else | a new array | | `a[1:4].copy()` | a range, detached | a copy | ## Common NumPy indexing and slicing mistakes - **Editing a slice you meant to keep separate.** Slices are views, so call `.copy()` first. - **Using `and`/`or` with arrays.** Use `&`, `|` and `~`, with parentheses around each comparison. - **Chaining fancy indexes on the left of `=`.** `d[[0, 2]][0] = 99` edits a temporary copy and leaves `d` unchanged; write `d[0] = 99` instead. - **Writing `b[1][2]` in hot loops.** It builds a temporary array; `b[1, 2]` does not. - **Assuming out-of-range slices raise errors.** `a[2:100]` quietly returns `[30 40 50]`, whereas `a[10]` raises `IndexError`. ## Related reading [NumPy Part 1: arrays](https://www.1stepgrow.com/articles/numpy-tutorial) covers the basics. [NumPy Part 3: broadcasting](https://www.1stepgrow.com/articles/numpy-broadcasting) continues the series, and [Part 4](https://www.1stepgrow.com/articles/numpy-aggregation-functions) finishes it with aggregation and linear algebra. Python's own slicing rules for sequences are covered in [lists vs tuples in Python](https://www.1stepgrow.com/articles/lists-vs-tuples-python). ## Frequently asked questions ### Why did modifying a slice change my original array? Basic slicing returns a view that shares the same memory buffer as the original array, not a copy. This is deliberate, because it makes slicing almost free even on very large arrays. The side effect is that any assignment through the slice writes into the source. Call .copy() on the slice whenever you need an independent array to modify. ### How do I check whether a NumPy array is a view? Look at the .base attribute. It is None for an array that owns its data, and it points at the source array for a view, so arr_slice.base is original returns True. For a more general test between any two arrays, np.shares_memory(a, b) tells you whether they overlap in memory, which also catches views of views. ### Why does mask1 and mask2 raise an error in NumPy? Python's and keyword needs a single True or False for the whole array, and NumPy refuses to guess, so it raises the familiar error that the truth value of an array is ambiguous. Use the elementwise operators &, | and ~ instead, and wrap each comparison in parentheses because those operators bind more tightly than > and <. ### What is the difference between fancy indexing and slicing? Slicing uses start:stop:step ranges and returns a view into the same memory. Fancy indexing uses an array or list of positions, or a boolean mask, and always returns a copy, because the selected elements are not generally evenly spaced in memory. Both can be used on the left of an assignment to change the original array in place. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/numpy-indexing-slicing/). Cite with the title, "1stepGrow" and a link._ --- # AI Engineer Roadmap 2026: From First API Call to Production Systems > A six-month AI engineer roadmap built on what teams hire for in 2026: retrieval, evaluation, cost control and guardrails, with projects and official resources. - **Author:** Althaf Ashraf — AI Systems Engineer, Tata Consultancy Services (https://www.1stepgrow.com/authors/althaf-ashraf/) - **Published:** Aug 5, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Artificial Intelligence · **Format:** Roadmap · **Read time:** 7 min - **Canonical URL:** https://www.1stepgrow.com/articles/ai-engineer-roadmap-2026/ ## Key takeaways - AI engineering is a software discipline. If you cannot write and test production Python, start there. - Evaluation is the skill that separates AI engineers from people who have used an API. Almost nobody teaches it. - Retrieval quality, not model choice, decides whether most AI products are useful. - Six months is realistic for a working engineer. For a non-programmer it is closer to fourteen. This AI engineer roadmap takes a working programmer from a first API call to a production AI system in about six months at 10–12 hours a week. It skips prompt tricks and spends its time on what interviews for the role actually probe: retrieval, evaluation, cost and safety. The common mistake is to treat AI engineering as a prompting skill. Teams hire people who can make a non-deterministic component behave inside a system with uptime commitments, a cost ceiling and users who type things nobody anticipated. That is a software engineering problem with an unusual dependency. This plan is written for developers who already ship code. If you do not program yet, it still applies, but budget closer to fourteen months. Each stage below has an exit test, official resources and a project you can defend in an interview. - **Total duration:** 6 months - **Weekly commitment:** 10-12 hours - **Typical entry salary:** Varies by market — see the pay section ### Stage 1: Production Python and the software basics (Weeks 1-4) Skip this only if you already ship code professionally. Everything downstream assumes you can structure a project, write tests and handle failure. **Skills:** Typed Python, Async & concurrency, Testing with pytest, API design, Git workflow, Docker basics **Projects:** - A small typed, tested service with a documented API **Outcome:** You can build a service another engineer would be willing to maintain. ### Stage 2: Foundation models — how they behave and why they fail (Weeks 5-8) Enough of the internals to reason about behaviour. Tokenisation, context windows, sampling, and the specific ways models fail that surprise people in production. **Skills:** Tokenisation & context limits, Sampling parameters, Structured output, Function/tool calling, Failure modes & hallucination **Projects:** - A tool-calling agent that reliably returns structured JSON against adversarial inputs **Outcome:** You can predict roughly how a model will fail before you deploy it, and design around it. ### Stage 3: Retrieval — where most AI products live or die (Weeks 9-14) The longest technical stage, because retrieval quality determines product quality far more than model choice does. Chunking, embeddings, hybrid search, reranking. **Skills:** Embeddings, Vector databases, Chunking strategy, Hybrid & keyword search, Reranking, Citation and grounding **Projects:** - A question-answering system over a real document corpus, with citations back to source - A documented comparison of three chunking strategies on the same corpus **Outcome:** You can explain why a retrieval system returns bad context and fix it methodically rather than by guessing. ### Stage 4: Evaluation — the actual differentiator (Weeks 15-18) Almost nobody teaches this, and it is the first thing a serious interviewer asks about. How do you know a change made the system better? **Skills:** Golden datasets, LLM-as-judge and its limits, Regression testing, Human evaluation design, Offline vs online metrics **Projects:** - An evaluation harness for your stage-three project that catches regressions automatically **Outcome:** You can prove a change improved the system, with numbers, rather than asserting it. ### Stage 5: Production concerns — cost, latency, safety (Weeks 19-22) The constraints that turn a demo into a product. Caching, streaming, fallbacks, rate limits, prompt injection defence and the cost model nobody budgets for. **Skills:** Caching & batching, Streaming responses, Fallback chains, Cost modelling, Prompt injection defence, Observability & tracing **Projects:** - Instrument your project with tracing, then cut its per-request cost by half without losing quality **Outcome:** You can run the system at a cost the business would accept, and explain the trade-offs you made. ### Stage 6: Ship it and write about it (Weeks 23-26) Deploy publicly, get real users, and publish what you learned. This stage is what turns six months of study into interviews. **Skills:** Deployment, Feedback loops, Technical writing, Interview narratives **Projects:** - A live product with real users and a public write-up of the architecture and its trade-offs **Outcome:** You have a URL, a set of numbers and a story — which is the entire technical interview. ## What does an AI engineer actually do? An AI engineer builds software products in which a language model is one component among many. The model sits behind retrieval, business logic, tests, logging and a budget. Most of the day-to-day work is integration. With [tool calling](https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview), the model returns a structured request and your application executes the function and sends the result back. Getting that loop right under bad input is engineering, not prompting. The role overlaps with two neighbours, and the clearest way to tell them apart is by what each one produces: | Role | What they mainly produce | Core skills | Evidence that gets them hired | |---|---|---|---| | AI engineer | Products built on existing models | Software engineering, retrieval, evaluation, cost control | A deployed system with measured quality | | Machine learning engineer | Trained models and training pipelines | ML theory, training, serving at scale | Models they trained and shipped | | Data scientist | Analyses and decisions | Statistics, SQL, experimentation | Analyses that changed a decision | If you are still choosing between these, read [Data Science vs Machine Learning vs AI](https://www.1stepgrow.com/articles/data-science-vs-machine-learning-vs-ai) before committing six months. ## How long does the AI engineer roadmap take, stage by stage? At 10–12 hours a week, the full plan is 26 weeks, or roughly 260–312 hours. The hours below are simply weeks multiplied by that weekly commitment. | Stage | Weeks | Approximate hours | Exit test | |---|---|---|---| | 1. Production Python | 1–4 | 40–48 | A typed, tested service another engineer could maintain | | 2. Model behaviour | 5–8 | 40–48 | Tool calls return valid JSON under adversarial input | | 3. Retrieval | 9–14 | 60–72 | You can explain why the wrong context came back | | 4. Evaluation | 15–18 | 40–48 | A regression suite fails when quality drops | | 5. Cost, latency, safety | 19–22 | 40–48 | Cost per request measured and reduced | | 6. Ship and write | 23–26 | 40–48 | A public URL, real users and a write-up | Six months assumes you already program. If you do not, stage one alone becomes several months, which is why a non-programmer should plan for something closer to fourteen. ## Which official resources should you use at each stage? Primary documentation beats course summaries here, because the tools change faster than courses are updated. As of September 2026, these are the references worth working through: - **Stage 1:** the Python [`typing` module docs](https://docs.python.org/3/library/typing.html) and [`asyncio` docs](https://docs.python.org/3/library/asyncio.html), the [pytest documentation](https://docs.pytest.org/en/stable/) and Docker's [getting started guide](https://docs.docker.com/get-started/). - **Stage 2:** your model provider's tool-use documentation. Anthropic's, for instance, supports `strict: true` on tool definitions so that tool calls match your JSON schema exactly. - **Stage 3:** the original [retrieval-augmented generation paper by Lewis et al. (2020)](https://arxiv.org/abs/2005.11401), then a practical store such as [pgvector](https://github.com/pgvector/pgvector), which adds exact and approximate nearest-neighbour search to PostgreSQL. - **Stage 4:** [Zheng et al. (2023)](https://arxiv.org/abs/2306.05685) on LLM evaluation with a model as judge. It found strong judges agreed with human preferences over 80% of the time, but also documented position, verbosity and self-enhancement biases. - **Stage 5:** the [OWASP Top 10 for LLM Applications 2025](https://genai.owasp.org/llmrisk/llm01-prompt-injection/), which ranks prompt injection first and covers indirect injection through websites and files, plus the [OpenTelemetry docs](https://opentelemetry.io/docs/) for tracing. ## Which skills get AI engineers hired? Interview loops for this role keep returning to the same three things, yet only one of them appears in most courses. **Retrieval quality.** Nearly every production AI application is a retrieval system with a language model attached. Candidates who can diagnose why the wrong context came back, and who have tried more than one chunking strategy on real data, stand out immediately. **Evaluation.** The question "how do you know that change made it better?" ends a surprising number of interviews. If your answer is "it looked better in a few examples", you are describing a hobby project. A golden dataset, a regression suite and honest error analysis are what a job looks like. **Cost and latency reasoning.** Someone has to answer what this costs at ten thousand users a day. Candidates who have instrumented a system and then actually reduced its cost have a story nobody else in the pipeline has. ## Which portfolio projects survive a technical screen? One deep project beats five shallow ones, because the interview will be a conversation about decisions. These four ideas each prove something specific: 1. **A support-ticket triager with tool calling.** It classifies tickets into a strict schema, and a test suite feeds it malformed, hostile and empty inputs. This proves you can make model output safe for downstream code. 2. **Cited question answering over documents you know well**, such as your employer's public policies or an open standard. Include a written comparison of chunking strategies with retrieval scores. 3. **An evaluation harness in CI.** A golden set of 50–100 questions, graded automatically, that fails the build when quality drops. Also record where your model-as-judge disagreed with your own grading. 4. **A cost-reduction write-up.** Trace a working system, then apply caching, smaller models for easy requests and shorter context, and publish the before-and-after numbers. For a worked example of the retrieval and evaluation pieces, see the guide to [building a RAG pipeline and testing it with RAGAS](https://www.1stepgrow.com/articles/rag-pipeline-and-ragas-guide). ## How do hiring teams screen AI engineer candidates? Most processes test general engineering first and AI-specific judgement second. A typical sequence looks like this: - **CV and portfolio screen.** A live link and a write-up with numbers get read; a list of frameworks does not. - **Coding round.** Usually ordinary software engineering, such as parsing, APIs and tests, rather than anything model-specific. - **System design with a model inside.** For example, "design a support assistant over our help centre", then probing on chunking, evaluation, fallbacks, prompt injection and cost. - **Project deep-dive.** Why you chose that chunk size, what broke, and what you would change. ## What does AI engineering pay? No government statistics agency tracks "AI engineer" as its own occupation, so every figure is a proxy. The table shows the most defensible ones. | Source | Population | Median annual pay | |---|---|---| | [US BLS, May 2025](https://www.bls.gov/ooh/computer-and-information-technology/software-developers.htm) | US software developers (all specialisms) | $135,980 | | [Stack Overflow Developer Survey 2025](https://survey.stackoverflow.co/2025/work) | AI/ML engineers, global, self-reported | $89,427 | | Stack Overflow Developer Survey 2025 | AI/ML engineers, US, self-reported | $189,500 | | Stack Overflow Developer Survey 2025 | AI/ML engineers, India, self-reported | $17,436 | Treat the survey figures with care, since respondents select themselves and the India sample is small, with 1,093 salary responses across all roles. No official Indian source publishes AI engineer salary bands, which is why the roadmap card above does not quote one. ## What should you skip on this path? Skip fine-tuning, framework hopping and leaderboard watching until you can evaluate a system. A lot of the advice attached to any AI engineer roadmap is noise, and these are the most common ways people lose months: - **Fine-tuning before you have exhausted retrieval and prompting.** It is expensive, it is usually the wrong tool, and interviewers know that. - **Framework tourism.** Learning four orchestration libraries teaches you four sets of abstractions over the same three ideas. - **Model leaderboards.** Chasing benchmark rankings is a hobby, because which model is marginally ahead this month rarely changes an architecture. - **Building agents before you can evaluate one.** An unevaluated agent is a demo that will embarrass you in production. - **Skipping tests because output varies.** Non-deterministic output still has deterministic contracts, such as valid JSON, required fields and refusal on out-of-scope requests, and you can test all of them. ## Where to go next Start with stage one's exit test this week: if you can already build a typed, tested service, move straight to stage two. If you are choosing between AI engineering and adjacent roles, read [Data Science vs Machine Learning vs AI](https://www.1stepgrow.com/articles/data-science-vs-machine-learning-vs-ai). The infrastructure side of the same work is covered in the [MLOps engineer career guide](https://www.1stepgrow.com/articles/mlops-engineer-career-guide), and [why prompt engineering stopped being a standalone career](https://www.1stepgrow.com/articles/prompt-engineering-career-reality) explains how this role absorbed it. For course options, our ranked list of the [best AI courses in 2026](https://www.1stepgrow.com/articles/best-ai-courses-2026) covers what each programme is genuinely good for. ## Frequently asked questions ### Do I need machine learning theory to be an AI engineer? Not to start. You need it to reach the senior end of the role, where 'the model is behaving strangely' becomes your problem rather than someone else's. Ship a working system first, then go back and learn the fundamentals of training and evaluation — that order keeps you motivated and gives the theory something concrete to attach to. ### Is AI engineering a real job or a temporary title? The title may not survive; the work will. Someone has to build retrieval, evaluate non-deterministic systems, defend against prompt injection and control inference cost. That work keeps expanding as more products put a language model inside them, whatever the job ends up being called on a careers page. ### Which framework should I learn? Learn the underlying concepts against the raw provider APIs first — tool calling, structured output, embeddings — and then adopt whichever framework your team uses. Frameworks in this space have a short half-life and change their abstractions often, while retrieval and evaluation fundamentals carry over from one stack to the next. ### Can I get an AI engineering job without a maths background? Yes, more easily than for machine learning engineering or data science. The role is dominated by systems thinking, testing discipline and product judgement. Maths becomes a ceiling later, when you need to reason about embeddings, fine-tuning or evaluation statistics in depth, but it is rarely a gate at entry. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/ai-engineer-roadmap-2026/). Cite with the title, "1stepGrow" and a link._ --- # NumPy Broadcasting in Python: Rules and Common Bugs (Part 3) > NumPy broadcasting explained: the two shape rules, why a (3,1) and a (4,) array give (3,4), and the silent bug that turns a vector sum into a matrix. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Aug 4, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/numpy-broadcasting/ ## Key takeaways - Broadcasting compares shapes from right to left, and each pair of dimensions must be equal or contain a 1. - The stretched operand is not copied; NumPy reuses the same values, which is why broadcasting is memory-efficient. - The classic bug is an (n, 1) column meeting an (n,) array and silently producing an (n, n) matrix. - Use np.newaxis, reshape or keepdims=True to control alignment deliberately rather than hoping. NumPy broadcasting is the set of rules that lets arithmetic work on arrays of different shapes, so you can write `arr + 10` instead of a loop. NumPy compares shapes from the right, and each pair of dimensions must be equal or contain a 1. Those two rules cover the whole system. The danger is that broadcasting can succeed when you wanted it to fail. Add a `(3,)` array to a `(3, 1)` array and you get a 3x3 matrix, not an error. This part of the series is for readers who know [indexing and slicing from Part 2](https://www.1stepgrow.com/articles/numpy-indexing-slicing). It works through the rules with shape diagrams, the silent bug, the tools that control alignment, and two worked examples. Outputs come from NumPy 2.2. ## What are the NumPy broadcasting rules? When NumPy operates on two arrays, it compares their shapes **element by element, starting from the trailing (rightmost) dimension**. According to the [official broadcasting guide](https://numpy.org/doc/stable/user/basics.broadcasting.html), two dimensions are compatible when: 1. they are equal, or 2. one of them is 1 If a dimension is missing from the shorter shape, it is treated as 1. If any pair fails both tests, then the operation raises `ValueError: operands could not be broadcast together`. Worked through: ```text A (3, 4) B (4,) ----------------- (3, 4) ✓ B is treated as (1,4), stretched down 3 rows ``` ```text A (3, 1) B (4,) ----------------- (3, 4) ✓ A stretched across, B stretched down ``` ```text A (3, 4) B (3,) ----------------- error ✗ 4 vs 3 in the trailing dimension ``` You can also ask NumPy to apply the rules without doing any arithmetic: ```python import numpy as np np.broadcast_shapes((3, 1), (4,)) # (3, 4) np.broadcast_shapes((8, 1, 6, 1), (7, 1, 5)) # (8, 7, 6, 5) ``` ## How does broadcasting work in practice? In practice, a scalar, a row or a column is stretched across a matrix so one operation applies everywhere. Scalar with array is the simplest broadcast: ```python a = np.arange(12).reshape(3, 4) a + 100 # every element gets 100 added a * 2 ``` Next, a row vector across a matrix: ```python matrix = np.array([[1, 2, 3], [4, 5, 6]]) # (2, 3) row = np.array([10, 20, 30]) # (3,) matrix + row # [[11 22 33] # [14 25 36]] ``` The row is applied to each row of the matrix. This is the pattern behind per-column operations such as centring: ```python data = np.array([[1., 2., 3.], [4., 5., 6.], [7., 8., 9.]]) col_means = data.mean(axis=0) # [4. 5. 6.], shape (3,) centred = data - col_means # each column gets its own mean subtracted ``` A column vector down a matrix, by contrast, needs an explicit second dimension: ```python col = np.array([[10], [20]]) # (2, 1) matrix + col # [[11 12 13] # [24 25 26]] ``` ## Why does NumPy broadcasting fail silently? Because shapes can be compatible in a way you did not intend. An `(n,)` array and an `(n, 1)` array both stretch, so NumPy returns an `(n, n)` matrix instead of raising an error, and nothing warns you. ```python a = np.array([1, 2, 3]) # shape (3,) b = np.array([[10], [20], [30]]) # shape (3, 1) a + b # [[11 12 13] # [21 22 23] # [31 32 33]] ``` You expected `[11 22 33]`. Instead, you got a 3x3 matrix. Right-aligning the shapes, `(3,)` becomes `(1, 3)` and meets `(3, 1)`. Both dimensions have a 1, so both stretch, and the result is the outer sum. Nothing errors. Downstream you get wrong numbers, or a memory explosion when the arrays are large: a `(10000, 1)` with a `(10000,)` produces a hundred-million-element array. **The fix is always to check shapes:** ```python print(a.shape, b.shape) # (3,) (3, 1) a + b.ravel() # [11 22 33] - flatten b to (3,) ``` The row-wise version of this bug raises an error instead, which is the lucky case: ```python M = np.array([[1., 2., 3.], [4., 5., 6.]]) M.mean(axis=1) # [2. 5.], shape (2,) M - M.mean(axis=1, keepdims=True) # [[-1. 0. 1.] [-1. 0. 1.]] ``` Without `keepdims=True`, `M - M.mean(axis=1)` fails with shapes `(2,3)` and `(2,)`, because the trailing dimensions 3 and 2 do not match. ## How do you control broadcasting with np.newaxis? Add a length-one dimension exactly where you want the stretch to happen, with `np.newaxis`, `reshape` or `keepdims=True`. The [NumPy indexing guide](https://numpy.org/doc/stable/user/basics.indexing.html) describes it as a way to insert a new axis of length one, and it is simply an alias for `None`: ```python a = np.array([1, 2, 3]) a[:, np.newaxis].shape # (3, 1) - column a[np.newaxis, :].shape # (1, 3) - row a.reshape(-1, 1).shape # (3, 1) - same as the first ``` This makes an outer operation explicit rather than accidental: ```python x = np.array([1, 2, 3]) y = np.array([10, 20]) x[np.newaxis, :] * y[:, np.newaxis] # [[10 20 30] # [20 40 60]] ``` Written that way, the intent is visible to whoever reads it next. ## Worked example: normalising features Broadcasting makes standard preprocessing a two-line operation: ```python rng = np.random.default_rng(0) X = rng.normal(size=(100, 5)) means = X.mean(axis=0) # (5,) stds = X.std(axis=0) # (5,) X_scaled = (X - means) / stds X_scaled.std(axis=0).round(10) # [1. 1. 1. 1. 1.] ``` Each of the 5 columns gets its own mean and standard deviation, applied down all 100 rows, with no loop and no copy of the statistics. The `axis` argument is the other half of this. `axis=0` collapses rows, giving one value per column, while `axis=1` collapses columns, giving one value per row. [Part 4](https://www.1stepgrow.com/articles/numpy-aggregation-functions) covers that argument in detail. ## Worked example: pairwise distances without a loop Adding a dimension to each copy of a point array turns every pair into one subtraction: ```python pts = np.array([[0, 0], [3, 4], [6, 8]]) # (3, 2) diff = pts[:, np.newaxis, :] - pts[np.newaxis, :, :] # (3, 3, 2) dist = np.sqrt((diff ** 2).sum(axis=-1)) # [[ 0. 5. 10.] # [ 5. 0. 5.] # [10. 5. 0.]] ``` The trade-off is memory. For `n` points the intermediate array is `n × n × 2`, so the NumPy documentation itself notes that broadcasting can be a bad idea when it creates a very large intermediate array. For tens of thousands of points, process in chunks instead. ## NumPy broadcasting quick reference | Shapes | Result | Why | |---|---|---| | `(3, 4)` and scalar | `(3, 4)` | scalar stretches everywhere | | `(3, 4)` and `(4,)` | `(3, 4)` | row applied to each row | | `(3, 4)` and `(3, 1)` | `(3, 4)` | column applied to each column | | `(3, 1)` and `(4,)` | `(3, 4)` | both stretch: an outer operation | | `(3, 4)` and `(3,)` | error | trailing 4 vs 3 | | `(2, 3)` and `(2,)` | error | trailing 3 vs 2; use `keepdims=True` | ## Common NumPy broadcasting mistakes - **Mixing `(n,)` and `(n, 1)`.** The sum becomes an `(n, n)` matrix; check `.shape` first. - **Dropping `keepdims=True` on row-wise statistics.** The result no longer lines up with the rows. - **Broadcasting in place into a smaller array.** `z = np.zeros(3); z += np.ones((3, 3))` raises "non-broadcastable output operand", because the result cannot fit in `z`. - **Building huge intermediates.** Pairwise tricks on large inputs can use more memory than a chunked loop. ## Related reading [NumPy Part 2: indexing and slicing](https://www.1stepgrow.com/articles/numpy-indexing-slicing) precedes this, and [NumPy Part 1](https://www.1stepgrow.com/articles/numpy-tutorial) starts the series. [NumPy Part 4: aggregation and linear algebra](https://www.1stepgrow.com/articles/numpy-aggregation-functions) continues. For the operators themselves, see [Python operators](https://www.1stepgrow.com/articles/python-operators). ## Frequently asked questions ### What is broadcasting in NumPy? Broadcasting is the set of rules that lets NumPy apply an arithmetic operation to arrays of different shapes. Shapes are compared from the trailing dimension, and any dimension of size 1 is virtually stretched to match the other array. No data is copied for the stretched operand, so adding a scalar or a row vector to a large matrix is both short to write and fast to run. ### Why did adding two NumPy vectors give me a 2D result? Almost certainly one was shaped (n, 1) and the other (n,). Right-aligned, those become (n, 1) and (1, n), and both stretch, so the result is an (n, n) table of every pairwise sum. Print .shape on both arrays before the operation, then flatten the column with ravel() or reshape the other array so the shapes match. ### Does broadcasting use extra memory? The stretched operand does not: NumPy steps through it with a stride of zero along the broadcast dimension instead of copying it. The result array is still allocated at full size, however. That is why an accidental (10000, 1) plus (10000,) operation can exhaust memory, because it creates a result with a hundred million elements. ### How do I add a dimension to a NumPy array? Use np.newaxis, which is simply an alias for None, inside the index. For a one-dimensional array of shape (3,), arr[:, np.newaxis] gives a (3, 1) column and arr[np.newaxis, :] gives a (1, 3) row. arr.reshape(-1, 1) produces the same column, and np.expand_dims(arr, axis=1) is a more explicit alternative if you prefer named functions. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/numpy-broadcasting/). Cite with the title, "1stepGrow" and a link._ --- # NumPy Aggregation, Sorting and Linear Algebra in Python (Part 4) > NumPy aggregation along axes, NaN-safe statistics, argmax, sorting and linalg.solve, plus why NumPy and pandas return different standard deviations. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Aug 3, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/numpy-aggregation-functions/ ## Key takeaways - axis=0 collapses rows and gives one value per column; axis=1 collapses columns. Getting this backwards is the most common NumPy mistake. - A single NaN poisons an entire aggregation, so use the nan-aware functions when your data has gaps. - argmin, argmax and argsort return positions, not values, which is how you look up related data. - Use @ for matrix multiplication and np.linalg.solve rather than inverting matrices. NumPy aggregation reduces an array to answers such as totals, means, maxima and positions, either for the whole array or along one axis with `axis=0` (per column) or `axis=1` (per row). The functions are simple. The defaults are where results go quietly wrong. One example: the same six numbers give a standard deviation of 1.708 in NumPy and 1.871 in pandas, because the two libraries use different divisors. This final part of the series is for readers who know [arrays](https://www.1stepgrow.com/articles/numpy-tutorial), [indexing](https://www.1stepgrow.com/articles/numpy-indexing-slicing) and [broadcasting](https://www.1stepgrow.com/articles/numpy-broadcasting). It covers axis reductions, NaN handling, positions, sorting and the linear algebra calls modelling work relies on. The outputs below come from NumPy 2.2 and pandas 2.3. ## How does NumPy aggregation work along an axis? Pass `axis` to name the dimension that collapses. With no axis, the reduction covers every element: ```python import numpy as np a = np.array([[1, 2, 3], [4, 5, 6]]) a.sum() # 21 - everything a.sum(axis=0) # [5 7 9] - one value per column a.sum(axis=1) # [ 6 15] - one value per row ``` The mental model that works: **the named axis disappears.** For example, an array of shape `(2, 3)` summed with `axis=0` loses the first dimension and returns shape `(3,)`. The same applies to every reduction: ```python a.mean(axis=0) # [2.5 3.5 4.5] a.max(axis=1) # [3 6] a.std() # 1.707825127659933 - population std by default a.std(ddof=1) # 1.8708286933869707 - sample std ``` That `ddof` default is a genuine trap. [NumPy's `std`](https://numpy.org/doc/stable/reference/generated/numpy.std.html) uses `ddof=0` (population), whereas [pandas' `std`](https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.std.html) uses `ddof=1` (sample). Consequently, the same six numbers give 1.708 in NumPy and 1.871 in pandas, which is a confusing afternoon if you do not know why. Keep dimensions when you need the result to broadcast back: ```python row_means = a.mean(axis=1, keepdims=True) # shape (2, 1) not (2,) a - row_means # [[-1. 0. 1.] [-1. 0. 1.]] ``` Without `keepdims`, this is exactly the [broadcasting bug](https://www.1stepgrow.com/articles/numpy-broadcasting) from part three. In a notebook, a single result such as `a.sum()` displays as `np.int64(21)`. That is how NumPy 2 represents scalars; `print()` shows `21`. ## How do you handle NaN in NumPy aggregation? Use the nan-aware functions such as `np.nanmean` and `np.nansum`, because NaN propagates and one missing value contaminates the whole result: ```python data = np.array([1.0, 2.0, np.nan, 4.0]) data.mean() # nan np.nanmean(data) # 2.3333333333333335 ``` The nan-aware family covers the common cases: ```python np.nansum(data) # 7.0 np.nanmean(data) # 2.3333333333333335 np.nanstd(data) # 1.247219128924647 np.nanmax(data) # 4.0 np.nanmedian(data) # 2.0 ``` Detecting and removing: ```python np.isnan(data) # [False False True False] data[~np.isnan(data)] # [1. 2. 4.] np.isnan(data).sum() # 1 - count of missing ``` Note that `np.nan == np.nan` is `False`, so equality never finds NaN. Always use `np.isnan` instead. Also, per the [`nanmean` documentation](https://numpy.org/doc/stable/reference/generated/numpy.nanmean.html), an all-NaN slice still returns NaN and raises a `RuntimeWarning`. ## How do you find where the maximum value is in NumPy? Use `argmax`, which returns the index rather than the value. `argmin` and `argmax` return indices rather than values, which is what you want when the array is aligned with something else: ```python temps = np.array([18, 25, 31, 12, 28]) cities = np.array(['Pune', 'Delhi', 'Nagpur', 'Shimla', 'Bhopal']) hottest = temps.argmax() # 2 cities[hottest] # 'Nagpur' ``` For 2D arrays, convert the flat index into coordinates: ```python grid = np.array([[1, 8], [3, 5]]) flat = grid.argmax() # 1 np.unravel_index(flat, grid.shape) # (0, 1) - shown as (np.int64(0), np.int64(1)) ``` Other useful position functions: ```python np.where(temps > 25) # (array([2, 4]),) - indices meeting a condition np.argsort(temps) # [3 0 1 4 2] - indices that would sort temps[np.argsort(temps)] # [12 18 25 28 31] ``` `argsort` is how you sort one array by another. For instance, `cities[np.argsort(temps)]` gives `['Shimla' 'Pune' 'Delhi' 'Bhopal' 'Nagpur']`, coldest first. ## How do you sort a NumPy array? Use `np.sort` for a sorted copy or the `.sort()` method to sort in place. Both sort along the last axis unless you pass `axis`: ```python a = np.array([[3, 1, 2], [9, 7, 8]]) np.sort(a) # [[1 2 3] [7 8 9]] - each row (last axis by default) np.sort(a, axis=0) # [[3 1 2] [9 7 8]] - each column a.sort() # in place, modifies a ``` `np.sort` returns a new array, whereas the `.sort()` method modifies in place and returns `None`. Assigning the result of `.sort()` to a variable is therefore a common and puzzling bug. ## Which NumPy linear algebra functions matter? In modelling work, the everyday set is `@` for matrix multiplication, `np.linalg.solve` for linear systems, `eigh` for symmetric matrices and `norm` for lengths: ```python A = np.array([[1., 2.], [3., 4.]]) b = np.array([5., 6.]) A @ A # [[ 7. 10.] [15. 22.]] - matrix multiplication A.T # transpose np.linalg.det(A) # -2.0000000000000004 np.linalg.matrix_rank(A) # 2 ``` The determinant shows a tiny floating-point error rather than exactly `-2.0`, which is normal. Compare floats with `np.allclose`, never with `==`. To solve `Ax = b`, use `solve` rather than inverting: ```python x = np.linalg.solve(A, b) # [-4. 4.5] np.allclose(A @ x, b) # True # not: np.linalg.inv(A) @ b - slower and less numerically stable ``` This matters more than it looks. Explicit inversion loses precision on ill-conditioned matrices, while [`np.linalg.solve`](https://numpy.org/doc/stable/reference/generated/numpy.linalg.solve.html) uses the LAPACK `gesv` routine, a factorisation that avoids forming the inverse. The advice holds in every numerical library, not just NumPy. For eigenvalues, choose the function by matrix type: ```python values, vectors = np.linalg.eig(A) # general square matrix S = np.array([[2., 1.], [1., 2.]]) values, vectors = np.linalg.eigh(S) # symmetric: values [1. 3.] ``` If you are implementing PCA by hand, a covariance matrix is symmetric, so use [`eigh`](https://numpy.org/doc/stable/reference/generated/numpy.linalg.eigh.html). It is faster and returns eigenvalues in ascending order, whereas `eig` guarantees no order. Finally, norms: ```python np.linalg.norm(b) # 7.810249675906654 - Euclidean length np.linalg.norm(A, axis=1) # [2.23606798 5. ] - row-wise norms ``` ## Worked example: z-scores and outliers This small example combines most of the series, computing per-feature z-scores and finding outliers: ```python rng = np.random.default_rng(0) X = rng.normal(loc=50, scale=10, size=(200, 4)) X[5, 2] = np.nan # inject a gap means = np.nanmean(X, axis=0, keepdims=True) stds = np.nanstd(X, axis=0, keepdims=True) z = (X - means) / stds # broadcasting outlier_rows = np.where(np.nanmax(np.abs(z), axis=1) > 3)[0] print(f"{len(outlier_rows)} rows contain an extreme value") # 4 rows contain an extreme value ``` It uses aggregation with an axis, NaN handling, `keepdims` for broadcasting, and `where` for positions: the whole series in eight lines. ## NumPy aggregation quick reference | Task | Function | NaN-safe version | |---|---|---| | Total | `np.sum` | `np.nansum` | | Average | `np.mean` | `np.nanmean` | | Middle value | `np.median` | `np.nanmedian` | | Spread | `np.std` (`ddof=0`) | `np.nanstd` | | Largest / smallest | `np.max`, `np.min` | `np.nanmax`, `np.nanmin` | | Position of largest | `np.argmax` | `np.nanargmax` | | Running total | `np.cumsum` | `np.nancumsum` | | Solve `Ax = b` | `np.linalg.solve` | not applicable | ## Common NumPy aggregation mistakes - **Swapping `axis=0` and `axis=1`.** Check the result's shape: the named axis should be gone. - **Comparing NumPy and pandas standard deviations.** Pass `ddof` explicitly so both use the same divisor. - **Finding NaN with `==`.** Use `np.isnan`, since NaN is not equal to itself. - **Writing `sorted_a = a.sort()`.** The method returns `None`; use `np.sort(a)` for a new array. - **Inverting to solve.** Use `np.linalg.solve`, and `eigh` for symmetric matrices. ## Related reading Start the series at [NumPy Part 1](https://www.1stepgrow.com/articles/numpy-tutorial), or revisit [Part 3 on broadcasting](https://www.1stepgrow.com/articles/numpy-broadcasting). For visualising the results, see the [Plotly tutorial](https://www.1stepgrow.com/articles/plotly-tutorial) and the [matplotlib styling guide](https://www.1stepgrow.com/articles/matplotlib-styling-customization). The [NumPy cheat sheet](https://www.1stepgrow.com/articles/numpy-cheat-sheet) summarises every function in one place. ## Frequently asked questions ### What is the difference between axis=0 and axis=1 in NumPy? The axis argument names the dimension being collapsed. axis=0 collapses down the rows, producing one result per column, while axis=1 collapses across the columns, producing one result per row. If you remember that the named axis disappears from the shape, the rest follows: a (2, 3) array summed with axis=0 returns shape (3,). ### Why does my NumPy mean return NaN? At least one element is NaN, and NaN propagates through arithmetic by design, so the whole mean becomes NaN. Use np.nanmean, np.nansum and their relatives to skip missing values instead. Be aware that if an entire row or column is NaN, np.nanmean still returns NaN for it and raises a RuntimeWarning about an empty slice. ### How do I find where the maximum is, not just what it is? np.argmax returns the index of the maximum rather than the value. For a multi-dimensional array it returns a flat index by default, so combine it with np.unravel_index(flat, arr.shape) to turn that into row and column coordinates. Alternatively, pass an axis, such as arr.argmax(axis=1), to get the position of the maximum in each row. ### Should I use np.linalg.inv to solve a linear system? No. np.linalg.solve(A, b) computes the solution directly with the LAPACK gesv routine, which is faster and numerically more stable than computing an inverse and multiplying. Explicit inversion loses precision on ill-conditioned matrices. Inverting a matrix is rarely the right operation in practice, even though textbooks often present the solution that way. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/numpy-aggregation-functions/). Cite with the title, "1stepGrow" and a link._ --- # Plotly Tutorial for Python: Basic Interactive Charts (Plotly Part 1) > Plotly tutorial for Python: interactive charts in one Plotly Express call, when to use Graph Objects, styling, faceting, and HTML exports 100x smaller. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Aug 2, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/plotly-tutorial/ ## Key takeaways - Plotly Express is the high-level API; use it for most charts and drop to Graph Objects only when you need control. - Interactivity (hover, zoom, pan) comes free and is the main reason to choose Plotly over matplotlib. - Plotly Express works best with tidy data: one row per observation, one column per variable. - fig.update_layout() is where most styling happens, and template='plotly_white' fixes the default look in one argument. This Plotly tutorial shows you how to make interactive charts in Python: install `plotly`, pass a DataFrame to one Plotly Express call such as `px.line`, and you get a chart you can hover over, zoom into and pan around. Where matplotlib produces static images, Plotly gives you that interactivity with no extra work. The defaults hide a few surprises. A blank figure in Jupyter often comes down to outdated Notebook packages, and the default HTML export of one chart below weighed 4.8 MB against under 40 KB with a single argument changed. This first part is for Python users who know a little pandas and want charts for screens: Express versus Graph Objects, the everyday chart types, styling, faceting and saving. As of September 2026 the current release is Plotly 7.1. The examples here were run on Plotly 6.3, and the [Plotly 7.0 release notes](https://github.com/plotly/plotly.py/releases/tag/v7.0.0) removes nothing they use. ## How do you install Plotly? Run `pip install plotly` (with pandas for the examples), and add `anywidget` if you work in Jupyter: ```bash pip install plotly pandas ``` If you use conda, `conda install -c conda-forge plotly` works too. For notebooks, the [Plotly getting-started page](https://plotly.com/python/getting-started/) lists `jupyterlab` and `anywidget` for JupyterLab, or `notebook>=7.0` and `anywidget` for the classic interface. If you still need a Python environment, the [Anaconda on Windows guide](https://www.1stepgrow.com/articles/install-anaconda-windows) covers the setup. ## What are Plotly Express and Graph Objects? Plotly has a high-level and a low-level interface, and knowing which you are using prevents most confusion. **Plotly Express** (`plotly.express`, conventionally `px`) builds a whole figure from a DataFrame in one call. It handles the large majority of charts. **Graph Objects** (`plotly.graph_objects`, conventionally `go`) constructs figures trace by trace. It is more verbose, but necessary when you need control Express does not expose. Express returns a Graph Objects figure, so you can start high-level and refine afterwards. That is the workflow to aim for. ```python import plotly.express as px import plotly.graph_objects as go import pandas as pd ``` ## How do you make basic charts with Plotly Express? Plotly Express works best with **tidy data**: one row per observation, one column per variable. ```python df = pd.DataFrame({ 'month': ['Jan', 'Feb', 'Mar', 'Apr', 'May'], 'revenue': [120, 145, 132, 178, 195], 'region': ['North', 'North', 'North', 'North', 'North'], }) fig = px.line(df, x='month', y='revenue', title='Monthly revenue') fig.show() ``` That is a complete interactive chart. Axis labels come from the column names, and hover and zoom work straight away. **Bar chart:** ```python fig = px.bar(df, x='month', y='revenue', title='Revenue by month', labels={'revenue': 'Revenue (₹k)', 'month': 'Month'}) fig.show() ``` **Scatter, with a third variable encoded as colour and a fourth as size:** ```python world = px.data.gapminder().query("year == 2007") # 142 countries fig = px.scatter( world, x='gdpPercap', y='lifeExp', size='pop', color='continent', hover_name='country', log_x=True, size_max=55, title='Life expectancy vs GDP per capita, 2007', ) fig.show() ``` `hover_name` is the small detail that makes Plotly worth using, because hovering a point names the country rather than showing bare coordinates. The gapminder sample data ships with Plotly, so this runs offline. **Histogram and box plot:** ```python px.histogram(world, x='lifeExp', nbins=30) px.box(world, x='continent', y='lifeExp', points='outliers') ``` ### Worked example: tidy data with several series When you have more than one region, keep the data long and let `color` split the lines: ```python sales = pd.DataFrame({ 'month': ['Jan', 'Feb', 'Mar'] * 2, 'region': ['North'] * 3 + ['South'] * 3, 'revenue': [120, 145, 132, 90, 110, 128], }) fig = px.line(sales, x='month', y='revenue', color='region', markers=True) fig.show() ``` Each region becomes its own trace with its own legend entry. Plotly Express also accepts wide data, for example `y=['north', 'south']` for two columns, but long data scales better once you add faceting. ## How do you style a Plotly chart? Use `update_layout` for titles, axes, templates and hover behaviour, and `update_traces` for lines and markers. Most adjustments go through `update_layout`: ```python fig = px.line(df, x='month', y='revenue') fig.update_layout( title={'text': 'Monthly revenue', 'x': 0.02, 'font': {'size': 20}}, xaxis_title='Month', yaxis_title='Revenue (₹ thousands)', template='plotly_white', hovermode='x unified', margin={'l': 60, 'r': 30, 't': 60, 'b': 50}, showlegend=False, ) fig.show() ``` Two of those are worth calling out. `template='plotly_white'` replaces the default grey background with something publishable. Meanwhile, `hovermode='x unified'` shows every series at the hovered x-position in one tooltip, which is almost always what you want on a multi-line chart. Trace-level styling uses `update_traces`: ```python fig.update_traces(line={'width': 3, 'color': '#2563eb'}, mode='lines+markers') ``` ## When should you drop to Graph Objects? When Express cannot express what you need, typically mixing chart types, build the figure directly: ```python fig = go.Figure() fig.add_trace(go.Bar( x=df['month'], y=df['revenue'], name='Revenue', marker_color='#2563eb', )) fig.add_trace(go.Scatter( x=df['month'], y=df['revenue'].rolling(2).mean(), name='2-month average', mode='lines+markers', line={'color': '#f59e0b', 'width': 3}, )) fig.update_layout(title='Revenue with trend', template='plotly_white') fig.show() ``` The pattern is always the same: create a `go.Figure()`, call `add_trace` for each series, then `update_layout`. The two-month average starts with a gap because the first month has no previous value to average with. ## How do you make small multiples in Plotly? Pass `facet_col` (and `facet_col_wrap`) to a Plotly Express function. Small multiples in one argument are Express's strongest feature: ```python gap = px.data.gapminder() fig = px.line( gap.query("continent == 'Asia'"), x='year', y='lifeExp', facet_col='country', facet_col_wrap=5, height=800, ) fig.update_yaxes(matches=None) # independent y-axes per panel fig.show() ``` The Asia subset has 33 countries, so this draws seven rows of panels. Doing that by hand in Graph Objects takes dozens of lines. The [facet plots guide](https://plotly.com/python/facet-plots/) covers `facet_row` and shared axes in more depth. ## How do you save a Plotly chart? Use `write_html` for an interactive file and `write_image` for PNG or PDF: ```python fig.write_html('chart.html') # interactive, self-contained fig.write_html('chart_cdn.html', include_plotlyjs='cdn') # small, loads Plotly.js online ``` The default HTML file embeds the whole Plotly.js library, so it is large (4.8 MB for the facet chart above on Plotly 6.3) but works offline. With `include_plotlyjs='cdn'` the same chart shrank to under 40 KB, although the reader then needs an internet connection. The [HTML export guide](https://plotly.com/python/interactive-html-export/) explains the other options. For a PNG or PDF, use `fig.write_image('chart.png', scale=2)`. According to the [static image export guide](https://plotly.com/python/static-image-export/), this needs Kaleido 1.0 or later and a Chrome or Chromium install on the machine. ## Plotly Express quick reference This table collects the functions used in this Plotly tutorial: | Chart | Function | Key arguments | |---|---|---| | Line | `px.line` | `x`, `y`, `color`, `markers=True` | | Bar | `px.bar` | `x`, `y`, `color`, `barmode` | | Scatter / bubble | `px.scatter` | `size`, `color`, `hover_name`, `log_x` | | Distribution | `px.histogram` | `x`, `nbins` | | Groups | `px.box` | `x`, `y`, `points` | | Small multiples | any of the above | `facet_col`, `facet_col_wrap` | ## Common mistakes in a first Plotly project - **Passing wide data with many series.** Reshape to long form with `pandas.melt`, then use `color`. - **Calling `write_image` without Kaleido or Chrome.** Install Kaleido 1.0+, then run `plotly_get_chrome` if no browser is found. - **Emailing 5 MB HTML files.** Use `include_plotlyjs='cdn'` when the reader will be online. - **Styling every chart by hand.** Set `template='plotly_white'` once, then change only what matters. - **Using an old Jupyter Notebook.** Plotly 6 and later need Notebook 7 or JupyterLab. ## What comes next in this Plotly tutorial? [Part two](https://www.1stepgrow.com/articles/advanced-plotly-charts) covers subplots, secondary axes, animation, 3D charts and annotation: the customisation that turns a working chart into a presentable one. ## Related reading [Plotly Part 2: advanced charts](https://www.1stepgrow.com/articles/advanced-plotly-charts) continues this series. For the static alternative, see [matplotlib styling and customisation](https://www.1stepgrow.com/articles/matplotlib-styling-customization). Plotly charts are built on arrays, so the [NumPy tutorial series](https://www.1stepgrow.com/articles/numpy-tutorial) is useful background. ## Frequently asked questions ### Should I use Plotly Express or Graph Objects? Start with Plotly Express. It builds a complete figure from a DataFrame in one call and returns a normal Graph Objects figure, so you can refine it afterwards. Move to Graph Objects when you need several trace types on one set of axes, fine-grained control over each trace, or a chart type that Express does not cover. ### Should I use Plotly or matplotlib? Use Plotly for anything read on a screen where hover and zoom help, such as dashboards, exploratory analysis and charts shared with colleagues. Use matplotlib for print, academic papers and precise static layouts. Many analysts use both: Plotly while exploring the data, and matplotlib when a figure has to go into a PDF or a slide at a fixed size. ### Do Plotly charts work in Jupyter? Yes, they render inline and stay interactive. Plotly's own instructions for JupyterLab are to install jupyterlab and anywidget in the same environment. For the classic interface you need Jupyter Notebook 7 or later plus anywidget, because Plotly 6 dropped support for Notebook 6. If a figure appears blank, check those package versions first. ### How do I save a Plotly chart as an image? fig.write_html('chart.html') keeps full interactivity and needs nothing extra. fig.write_image('chart.png') produces a static image but requires Kaleido 1.0 or later, which in turn needs Chrome or Chromium on the machine. If Chrome is missing, Plotly provides the plotly_get_chrome command to install a copy it can use. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/plotly-tutorial/). Cite with the title, "1stepGrow" and a link._ --- # Data Scientist Roadmap 2026: A Seven-Month Path from SQL to Offer > A stage-gated data scientist roadmap for 2026: SQL, statistics, modelling and communication, weighted as interviews test them, with projects and free resources. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Aug 1, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Data Science · **Format:** Roadmap · **Read time:** 7 min - **Canonical URL:** https://www.1stepgrow.com/articles/data-scientist-roadmap-2026/ ## Key takeaways - SQL is tested in almost every data science interview and is one of the most common reasons capable candidates fail early rounds. - Statistics matters more than modelling. Most real work is experiment design and inference, not tuning gradient boosters. - A portfolio of analyses that changed a decision beats a portfolio of Kaggle notebooks. - Communication is a graded interview competency, not a soft skill footnote. This data scientist roadmap takes you from SQL to a job offer in about seven months at 10–14 hours a week. It follows the order interviews test: can you get the data, can you tell whether a result is real, can you model it responsibly, and can you explain it to someone who will act on it. Most self-study plans over-weight the third and neglect the first and fourth. That mismatch is why capable people fail interviews they should pass, so this roadmap is weighted the way the interviews are. It is written for analysts, graduates and career changers who can give it 10–14 hours a week. Each of the six stages ends in an exit test, so you know when to move on. The sections after the roadmap add hours per stage, free official resources, portfolio projects that work and what official pay data actually shows. - **Total duration:** 7 months - **Weekly commitment:** 10-14 hours - **Typical entry salary:** Varies by market — see the pay section ### Stage 1: SQL until it is boring (Weeks 1-5) The most under-rated stage. SQL appears in nearly every data science interview and is where most candidates are quietly filtered out in round one. **Skills:** Joins & set operations, Window functions, CTEs & subqueries, Aggregation logic, Query performance basics **Projects:** - Fifty medium and hard SQL problems solved without hints - One analysis answered entirely in SQL, no pandas **Outcome:** You can write a window-function query under observation without freezing. ### Stage 2: Statistics that survives scrutiny (Weeks 6-12) Inference, not just descriptive summaries. The ability to say whether a result is real, and how confident you are, is the core of the job. **Skills:** Distributions & sampling, Hypothesis testing, Confidence intervals, A/B test design, Power analysis, Common statistical traps **Projects:** - Design and analyse a simulated A/B test, including power calculation and a written recommendation **Outcome:** You can explain why a result is or is not significant to someone who wants it to be significant. ### Stage 3: Python for analysis (Weeks 13-17) Pandas, visualisation and reproducible analysis. Focus on clarity and correctness rather than clever one-liners. **Skills:** pandas, NumPy, matplotlib/seaborn, Notebook hygiene, Reproducible pipelines **Projects:** - An end-to-end exploratory analysis of a messy public dataset, with data quality issues documented **Outcome:** You can take a raw, dirty dataset to a defensible conclusion and someone else can rerun your work. ### Stage 4: Modelling with judgement (Weeks 18-24) Regression, classification, trees and the evaluation discipline around them. The hard part is choosing the metric and knowing when a model is the wrong answer. **Skills:** Linear & logistic regression, Tree ensembles, Cross-validation, Metric selection, Feature engineering, Leakage detection **Projects:** - A predictive model with honest validation, an error analysis section, and a stated business threshold **Outcome:** You can justify your metric choice and identify leakage before a reviewer does. ### Stage 5: Communication as a graded skill (Weeks 25-28) The stage that separates offers from rejections at the final round. Written memos, verbal explanation and stakeholder pushback. **Skills:** Analytical writing, Data visualisation for decisions, Executive summaries, Handling disagreement **Projects:** - A one-page decision memo from your stage-four project, written for a non-technical reader - A ten-minute recorded walkthrough of an analysis **Outcome:** A non-technical person can read your memo and know what to do differently. ### Stage 6: Interview and apply (Weeks 29-32) Case studies, SQL under time pressure, product sense and the behavioural round. All practised deliberately, not improvised. **Skills:** Case interview practice, Product sense questions, Timed SQL, Behavioural stories **Projects:** - Eight mock interviews covering SQL, stats, case and behavioural **Outcome:** You can hold a case interview conversation without needing the answer to be right immediately. ## How long does each stage of the data scientist roadmap take? The whole plan runs 32 weeks, which at 10–14 hours a week is roughly 320–450 hours. The hours below are simply weeks multiplied by that range. | Stage | Weeks | Approximate hours | You are ready to move on when… | |---|---|---|---| | 1. SQL | 1–5 | 50–70 | You write window-function queries under observation | | 2. Statistics | 6–12 | 70–98 | You can design an A/B test, including sample size | | 3. Python for analysis | 13–17 | 50–70 | Someone else can rerun your analysis end to end | | 4. Modelling | 18–24 | 70–98 | You can justify your metric and spot leakage | | 5. Communication | 25–28 | 40–56 | A non-technical reader acts on your memo | | 6. Interviewing | 29–32 | 40–56 | Eight mock interviews done, weak rounds fixed | If you already work as an analyst, stages one and three may take half the time. However, do not skip the exit tests, because they are what the interview will check. ## Why does the SQL round stop so many candidates? If you take one thing from this page, take this: a common point of failure in data science hiring is a live SQL round early in the process. Not because the questions are exotic. Rather, candidates practised modelling for six months and SQL for six days, then had to write a window function while a stranger watched. SQL is also simply everywhere. In the [Stack Overflow Developer Survey 2025](https://survey.stackoverflow.co/2025/technology), 58.6% of all respondents said they had done extensive work in SQL over the past year, slightly more than Python at 57.9%. Five weeks at the start of this roadmap looks disproportionate. It is not. In fact, it is the stage with the highest return per hour in the entire path. ## Which free, official resources cover each stage? You do not need a paid course for any single stage, although you do need a deadline and feedback. As of September 2026, these primary resources cover the ground: - **SQL:** the PostgreSQL manual's [tutorial on window functions](https://www.postgresql.org/docs/current/tutorial-window.html), which is short, precise and exactly what the interview round tests. - **Statistics for data science:** [OpenIntro Statistics](https://www.openintro.org/book/os/), a free PDF textbook covering inference, categorical data and regression. - **A/B testing and power:** the power and sample-size tools in [statsmodels](https://www.statsmodels.org/stable/stats.html), such as `TTestIndPower`, so that your test design includes a sample-size calculation rather than a guess. - **Python for analysis:** the pandas [10 minutes to pandas](https://pandas.pydata.org/docs/user_guide/10min.html) guide, then the rest of the user guide as you need it. - **Modelling:** scikit-learn's [common pitfalls page](https://scikit-learn.org/stable/common_pitfalls.html). It defines data leakage and recommends splitting data before any preprocessing and using pipelines. For deeper SQL practice with the reasoning written out, work through these [SQL interview questions and answers](https://www.1stepgrow.com/articles/sql-interview-questions-answers). ## What should a data science portfolio contain? Two or three analyses that each end in a recommendation, rather than a folder of modelling notebooks. The portfolio advice for data science is different from software engineering, and people get it wrong by copying the wrong template. A software portfolio shows you can build. A data science portfolio needs to show you can *decide*. That means the artefact is not really the notebook — it is the memo at the end that says: here is what I found, here is my confidence in it, here is what I would do, and here is what would change my mind. Two well-scoped analyses with clear recommendations will out-perform ten polished modelling notebooks in almost every hiring conversation. Some ideas that work well: 1. **An experiment readout.** Simulate or find an A/B test dataset, compute the power the test needed, then write a recommendation that includes what you would *not* conclude. 2. **A messy public dataset, cleaned honestly.** For example, local government or transport data, with every data quality problem logged and its effect on the conclusion stated. 3. **A churn or demand model with a business threshold.** Choose the metric from the cost of errors, show the leakage checks you ran, and include an error analysis. 4. **A decision memo.** One page, written for a manager, based on one of the projects above. ## How do data science interviews actually screen? Most processes test the same competencies in a fairly stable order, even though the formats vary by company: - **Screening call and CV review.** They look for evidence of analyses, not tool lists. - **SQL round.** Usually live or timed, with joins, aggregation and window functions. - **Statistics and experimentation.** Hypothesis tests, confidence intervals, A/B test design and common traps such as peeking. - **Case or product sense.** "Sign-ups fell 10% last week — what do you check first?" - **Modelling discussion.** Metric choice, validation and leakage, often built around your own project. - **Behavioural round.** Disagreement, ambiguity and influence without authority. The round-by-round breakdown in [data science interview rounds](https://www.1stepgrow.com/articles/data-science-interview-rounds) goes further on each format. ## What does a data scientist earn? The cleanest official figures are American. The [US Bureau of Labor Statistics](https://www.bls.gov/ooh/math/data-scientists.htm) reports a median annual wage of $120,230 for data scientists in May 2025, with 275,600 jobs in 2025, employment projected to grow 35% from 2025 to 2035, and about 24,800 openings a year. | Source | Population | Median annual pay | |---|---|---| | US BLS, May 2025 | US data scientists | $120,230 | | [Stack Overflow Developer Survey 2025](https://survey.stackoverflow.co/2025/work) | Data scientists, global, self-reported | $82,910 | | Stack Overflow Developer Survey 2025 | Data scientists, US, self-reported | $145,000 | | Stack Overflow Developer Survey 2025 | Data scientists, India, self-reported | $11,624 | Read the survey rows as a rough guide only, because respondents choose themselves and the India sample is small. No official Indian source publishes data scientist salary bands, which is why the roadmap card does not quote one. For more on what the official figures do and do not show, see [data science salary by role](https://www.1stepgrow.com/articles/data-science-salary-by-role). ## Which mistakes stall a data scientist roadmap? Most people who stall on this path make one of a handful of avoidable errors: - **Starting with deep learning.** It feels like progress, yet it is rarely what an entry-level interview tests. - **Treating statistics as a formula sheet.** Interviewers ask *why* a test applies, not only how to run it. - **Leaking the target.** Scaling or imputing on the full dataset before splitting inflates every score, which a reviewer will catch. - **Portfolios with no conclusion.** A notebook that ends in a confusion matrix, rather than a recommendation, shows method but not judgement. - **Applying before the SQL stage is solid.** An early SQL failure can close a company's door for months. ## Related reading Still choosing a track? Start with [Data Science vs Machine Learning vs AI](https://www.1stepgrow.com/articles/data-science-vs-machine-learning-vs-ai), or the narrower [Data Analyst vs Data Scientist](https://www.1stepgrow.com/articles/data-analyst-vs-data-scientist) comparison. For course options, see the [best data science courses in India](https://www.1stepgrow.com/articles/best-data-science-courses-india), and before paying for a structured programme, read [is a data science bootcamp worth it](https://www.1stepgrow.com/articles/is-a-data-science-bootcamp-worth-it). ## Frequently asked questions ### Do I need a master's degree to become a data scientist? In research-heavy teams and some large enterprises, it still functions as a filter. In product companies, a portfolio of analyses that changed decisions carries more weight. It helps; it is not a prerequisite outside specific sectors. The US Bureau of Labor Statistics lists a bachelor's degree as the typical entry-level education for the occupation. ### How important is SQL really? More important than any other single technical skill on this list. It is tested in nearly every hiring process, often first, and it is where a large share of otherwise strong candidates are eliminated. Window functions, joins that change row counts and aggregation logic are the usual stumbling points, so practise them under a timer. ### Are Kaggle competitions good portfolio pieces? As evidence of modelling skill, yes. As evidence you can do the job, only partly — competitions hand you a clean dataset and a fixed metric, which is the part of real work that does not exist. Pair one with an analysis you scoped yourself, including the question, the data cleaning and a written recommendation. ### Is data science still a growing field? In the United States, the Bureau of Labor Statistics projects data scientist employment to grow 35% between 2025 and 2035, much faster than average. Within the field, routine reporting is being absorbed by better tooling, while experimentation, causal inference and decision support are where the demanding work sits. Aim your learning at the second group. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/data-scientist-roadmap-2026/). Cite with the title, "1stepGrow" and a link._ --- # Advanced Plotly Charts: Subplots, Animation and 3D (Plotly Part 2) > Advanced Plotly charts in Python: subplots, dual axes, animation, 3D and heatmaps, plus the axis-range mistake that silently drops points from animations. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Jul 31, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/advanced-plotly-charts/ ## Key takeaways - make_subplots is the entry point for multi-panel figures; pass specs when you need a secondary y-axis. - Animation is a single argument in Plotly Express, but fixed axis ranges are what make the motion readable. - 3D charts are usually worse than a good 2D chart, so use them only when the third dimension carries real information. - An annotation that names the insight does more for comprehension than any styling change. Advanced Plotly charts are the tools for when one line or bar is not enough: `make_subplots` for several panels, `secondary_y` for two scales, `animation_frame` for change over time, and annotations for a chart that explains itself. Each is a few lines of code. The hard part is knowing when each one misleads. Take animation. Copy the axis ranges from Plotly's own gapminder example and Plotly draws four data points (Kuwait in three years, Rwanda in one) off the chart, with no warning. This part of the series is for readers who already know Plotly Express from [part one](https://www.1stepgrow.com/articles/plotly-tutorial). You get runnable code for subplots, dual axes, animation, 3D, heatmaps and annotation, plus a table for picking the right one. Every example below was run on Plotly 6.3. As of September 2026 the current release is Plotly 7.1, and nothing used here was removed in 7.0. ## How do you build subplots in Plotly? Multi-panel figures need `make_subplots` from `plotly.subplots`, as the [subplots guide](https://plotly.com/python/subplots/) shows: ```python import plotly.graph_objects as go from plotly.subplots import make_subplots import numpy as np x = np.linspace(0, 10, 100) fig = make_subplots( rows=2, cols=2, subplot_titles=('Sine', 'Cosine', 'Damped', 'Noise'), horizontal_spacing=0.10, vertical_spacing=0.14, ) fig.add_trace(go.Scatter(x=x, y=np.sin(x), name='sin'), row=1, col=1) fig.add_trace(go.Scatter(x=x, y=np.cos(x), name='cos'), row=1, col=2) fig.add_trace(go.Scatter(x=x, y=np.sin(x) * np.exp(-x / 5), name='damped'), row=2, col=1) rng = np.random.default_rng(0) fig.add_trace(go.Scatter(x=x, y=rng.normal(size=100), mode='markers', name='noise'), row=2, col=2) fig.update_layout(height=600, showlegend=False, template='plotly_white', title_text='Four panels') fig.show() ``` Axis updates also take the same `row`/`col` arguments: ```python fig.update_yaxes(title_text='Amplitude', row=1, col=1) fig.update_xaxes(title_text='Time (s)', row=2, col=1) ``` If stacked panels share a time axis, pass `shared_xaxes=True` to `make_subplots`. Panels in the same column are then linked, so zooming one zooms the others, and only the bottom panel keeps its x tick labels. ## When should you use a secondary y-axis? Use one only for two series on genuinely different scales. The [multiple axes guide](https://plotly.com/python/multiple-axes/) uses the same pattern: ```python fig = make_subplots(specs=[[{'secondary_y': True}]]) months = ['Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun'] revenue = [120, 145, 132, 178, 195, 210] margin = [0.22, 0.24, 0.21, 0.27, 0.29, 0.28] fig.add_trace(go.Bar(x=months, y=revenue, name='Revenue', marker_color='#2563eb'), secondary_y=False) fig.add_trace(go.Scatter(x=months, y=margin, name='Margin', mode='lines+markers', line={'color': '#f59e0b', 'width': 3}), secondary_y=True) fig.update_yaxes(title_text='Revenue (₹k)', secondary_y=False) fig.update_yaxes(title_text='Margin', tickformat='.0%', secondary_y=True) fig.update_layout(template='plotly_white', hovermode='x unified') fig.show() ``` A caution worth stating: dual axes let you make any two series look correlated by choosing the scales. Therefore, use them when the units really differ and the comparison is legitimate, and be suspicious when you see them elsewhere. ## How do you animate a Plotly chart without the axes jumping? Pass `animation_frame` to a Plotly Express function and fix `range_x` and `range_y` yourself. The single argument creates the animation; the fixed ranges are what make it readable: ```python import plotly.express as px gap = px.data.gapminder() fig = px.scatter( gap, x='gdpPercap', y='lifeExp', size='pop', color='continent', hover_name='country', animation_frame='year', animation_group='country', log_x=True, size_max=55, range_x=[100, 150_000], range_y=[20, 90], title='Development over time', ) fig.show() ``` `animation_group` keeps each country's point identified across frames, so it moves rather than disappearing and reappearing. The explicit `range_x` and `range_y` matter just as much: without them the axes rescale each frame and the motion becomes meaningless. ### Worked example: choosing the ranges The [Plotly animation guide](https://plotly.com/python/animations/) advises always fixing the ranges so the data stays visible, and its own example uses `[100, 100000]` and `[25, 90]`. However, check your data first: ```python gap['gdpPercap'].max() # 113523.1329 - Kuwait, 1957 gap['lifeExp'].min() # 23.599 - Rwanda, 1992 ``` Both values fall outside those example ranges, so three Kuwait frames and one Rwanda frame would be drawn off the chart. That is why the example above uses the wider `[100, 150_000]` and `[20, 90]`, which cover all 12 frames from 1952 to 2007. ## When is a 3D Plotly chart worth using? Only when the third dimension is the message, such as the surface of a mathematical function or genuinely spatial data. For everything else, a 2D chart with colour, size or facets reads better: ```python x = np.linspace(-5, 5, 60) y = np.linspace(-5, 5, 60) X, Y = np.meshgrid(x, y) Z = np.sin(np.sqrt(X**2 + Y**2)) fig = go.Figure(go.Surface(x=X, y=Y, z=Z, colorscale='Viridis')) fig.update_layout( title='A surface', scene={'xaxis_title': 'x', 'yaxis_title': 'y', 'zaxis_title': 'z'}, height=600, ) fig.show() ``` Surfaces of a genuine mathematical function are a fair use, because the shape is the point. By contrast, a 3D bar chart of categorical data is almost never better than a heatmap or a faceted 2D chart, because the perspective distorts comparison and near objects hide far ones. ## Why is annotation the most useful advanced Plotly feature? A chart that states its own point is worth several that leave the reader to find it, so this is where extra effort pays off first. ```python fig = go.Figure(go.Scatter(x=months, y=revenue, mode='lines+markers', line={'color': '#2563eb', 'width': 3})) fig.add_annotation( x='Apr', y=178, text='Pricing change
launched', showarrow=True, arrowhead=2, ax=-50, ay=-60, bgcolor='white', bordercolor='#cbd5e1', borderwidth=1, borderpad=6, ) fig.add_vrect(x0='Apr', x1='Jun', fillcolor='#2563eb', opacity=0.06, line_width=0, annotation_text='Post-change') fig.add_hline(y=150, line_dash='dash', line_color='#94a3b8', annotation_text='Target') fig.update_layout(template='plotly_white', title='Revenue, with context') fig.show() ``` `add_vrect`, `add_hline` and `add_annotation` between them cover most of what you need to make a chart self-explanatory. The [shapes guide](https://plotly.com/python/horizontal-vertical-shapes/) lists the matching `add_vline` and `add_hrect`, plus `annotation_position` for placing the label. On a category axis like this one, the month names work directly as positions. ## How do you make a correlation heatmap in Plotly? Use `px.imshow` on a correlation matrix, with `zmin=-1` and `zmax=1` so the neutral colour means zero. It is also the usual answer when you were reaching for 3D: ```python corr = gap[['lifeExp', 'pop', 'gdpPercap', 'year']].corr() fig = px.imshow( corr, text_auto='.2f', color_continuous_scale='RdBu_r', zmin=-1, zmax=1, title='Correlation matrix', ) fig.show() ``` On the gapminder data, the strongest pairing is life expectancy with GDP per capita, at 0.58. Setting `zmin` and `zmax` symmetrically around zero matters for diverging scales; otherwise, the colour midpoint drifts and the chart misleads. The `text_auto` format string follows d3-format, as the [heatmaps guide](https://plotly.com/python/heatmaps/) notes. ## Which advanced Plotly chart should you use? | You want to show | Use | Avoid | |---|---|---| | Different charts side by side | `make_subplots` | several separate figures | | The same chart per category | `facet_col` in Plotly Express | hand-built subplots | | Two units on one timeline | `secondary_y=True` | dual axes for same-unit data | | Change over time | `animation_frame` with fixed ranges | auto-scaling axes | | A third numeric variable | colour, size or a heatmap | 3D bars | | A mathematical surface | `go.Surface` | a flat contour if shape matters | | The takeaway | `add_annotation`, `add_vrect`, `add_hline` | a long caption | ## Common mistakes with advanced Plotly charts - **Forgetting `row` and `col`.** Traces added without them do not go where you intended, so always pass both. - **Leaving animation ranges automatic.** The axes jump between frames; fix both ranges. - **Copying example ranges without checking the data.** Points outside the range silently vanish. - **Using a diverging colour scale without `zmin`/`zmax`.** The neutral colour stops meaning zero. - **Styling before annotating.** One clear annotation usually matters more than colour tweaks. ## A practical default for presentable charts Build with Express, style lightly, annotate once. In practice, the sequence that produces good advanced Plotly charts quickly is this: first build the chart with Express, then set `template='plotly_white'` and `hovermode='x unified'` on anything time-based. Finally, add one annotation naming the insight, and stop there, since further styling has rapidly diminishing returns. ## Related reading [Plotly Part 1](https://www.1stepgrow.com/articles/plotly-tutorial) covers the basics. For static, print-quality output, see [matplotlib styling and customisation](https://www.1stepgrow.com/articles/matplotlib-styling-customization). The arrays behind these examples are explained in [NumPy Part 3 on broadcasting](https://www.1stepgrow.com/articles/numpy-broadcasting), and the planned guide to [Plotly maps, treemaps and 3D plots](https://www.1stepgrow.com/articles/plotly-maps-treemaps-3d-plots) goes further into specialist charts. ## Frequently asked questions ### How do I put two charts side by side in Plotly? Use plotly.subplots.make_subplots(rows=1, cols=2), then add each trace with the row and col arguments. Plotly Express cannot merge two finished figures into one, so arbitrary multi-panel work generally means Graph Objects. The exception is small multiples of the same chart, where facet_col in Plotly Express is far shorter than building subplots by hand. ### How do I add a second y-axis in Plotly? Pass specs=[[{'secondary_y': True}]] to make_subplots, then add each trace with secondary_y=True or secondary_y=False, and title each axis with update_yaxes using the same argument. Use it sparingly, because dual axes make it easy to imply a relationship that is not there simply by choosing the two scales. ### Are 3D plots in Plotly useful? Rarely. Occlusion and perspective make values hard to compare, and a static screenshot loses the rotation that made the chart readable on screen. Prefer colour, size or faceting to encode a third variable. Keep 3D for genuinely spatial data or for surfaces of a mathematical function, where the shape itself is the message. ### Why does my Plotly animation look wrong? Usually the axes are rescaling on every frame, so points appear to jump even when the data barely changes. Set range_x and range_y wide enough to cover every frame. Also pass animation_group so each entity keeps its identity between frames and moves smoothly instead of disappearing and reappearing. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/advanced-plotly-charts/). Cite with the title, "1stepGrow" and a link._ --- # Advanced Matplotlib Styling and Customisation in Python (Part 5) > Matplotlib styling for report-ready charts: rcParams, style sheets, colormaps, annotation and clean export, and why plt.style.use('seaborn') now fails. - **Author:** Vanshika Nigam — ETL & Data Engineer, Accenture (https://www.1stepgrow.com/authors/vanshika-nigam/) - **Published:** Jul 30, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/matplotlib-styling-customization/ ## Key takeaways - Use the object-oriented API (fig, ax) rather than the pyplot state machine, because it is the reliable way to control complex figures. - rcParams set once at the top of a notebook beat repeating the same styling on every chart. - Removing chart junk, such as the top and right spines and heavy grids, improves almost every default matplotlib figure. - Direct labelling usually beats a legend, and one annotation that names the insight beats any colour tweak. Matplotlib styling is the gap between a chart that works and one you would put in a report. The fix is mostly configuration: set `rcParams` once, pick a style sheet, choose colours deliberately, annotate the insight and export with `bbox_inches='tight'`. A handful of those settings, mainly the spines and grid, does most of the work. Old tutorials make this harder than it should be. Their `plt.style.use('seaborn')` now raises an `OSError`, and their rainbow colormaps invent boundaries that are not in the data. This part is for readers who can already produce basic plots and want control: the object-oriented API, defaults, colour, annotation, direct labels and export, each with code run on a recent release. The code was run on Matplotlib 3.10. As of September 2026 the current release is 3.11, and every style name and function used here is in the [3.11 documentation](https://matplotlib.org/stable/users/explain/customizing.html). ## Why use the object-oriented API? Matplotlib has two interfaces. The `pyplot` state machine (`plt.plot`, `plt.title`) draws on whichever Axes is currently active. It is convenient for one chart, but unreliable for anything more. The object-oriented API, by contrast, gives you explicit handles. Matplotlib's guide to its [application interfaces](https://matplotlib.org/stable/users/explain/figure/api_interfaces.html) notes that complicated plots often end up simpler this way. ```python import matplotlib.pyplot as plt import numpy as np x = np.linspace(0, 10, 200) fig, ax = plt.subplots(figsize=(8, 4.5)) ax.plot(x, np.sin(x), label='sin(x)') ax.plot(x, np.cos(x), label='cos(x)') ax.set_xlabel('x') ax.set_ylabel('amplitude') ax.set_title('Trigonometric functions') ax.legend() plt.show() ``` `fig` is the whole canvas, while `ax` is one plotting area. Every customisation below hangs off one of those two objects. For grids of charts: ```python fig, axes = plt.subplots(2, 2, figsize=(10, 7), sharex=True) axes[0, 0].plot(x, np.sin(x)) axes[0, 1].plot(x, np.cos(x)) axes[1, 0].hist(np.random.default_rng(0).normal(size=500), bins=30) axes[1, 1].scatter(x[::5], np.sin(x[::5])) fig.suptitle('Four panels', fontsize=14) fig.tight_layout() ``` `sharex=True` links the axes, so setting limits on one applies across panels, and it also removes duplicated tick labels. As an alternative to `tight_layout`, you can pass `layout='constrained'` to `plt.subplots`. ## How do you set matplotlib styling defaults once? Rather than styling every chart, set `rcParams` at the top of the notebook: ```python plt.rcParams.update({ 'figure.figsize': (8, 4.5), 'figure.dpi': 110, 'savefig.dpi': 300, 'savefig.bbox': 'tight', 'font.size': 11, 'axes.titlesize': 14, 'axes.titleweight': 'bold', 'axes.labelsize': 11, 'axes.spines.top': False, 'axes.spines.right': False, 'axes.grid': True, 'grid.alpha': 0.25, 'grid.linestyle': '--', 'legend.frameon': False, 'lines.linewidth': 2, }) ``` Those settings alone transform the default look. In particular, the spine removal and the faint dashed grid do most of the work. Built-in style sheets are a quicker route: ```python print(plt.style.available) plt.style.use('seaborn-v0_8-whitegrid') ``` Note the `seaborn-v0_8-` prefix. Older tutorials use `plt.style.use('seaborn')`, but that name no longer exists, and current Matplotlib raises `OSError: 'seaborn' is not a valid package style`. The [style sheets reference](https://matplotlib.org/stable/gallery/style_sheets/style_sheets_reference.html) shows every current name. Styles can also be combined, with later styles overriding earlier ones, and applied temporarily with a context manager so they do not leak into other charts: ```python with plt.style.context('dark_background'): fig, ax = plt.subplots() ax.plot(x, np.sin(x)) ``` ## How do you choose colours in Matplotlib? Set your palette once through `axes.prop_cycle`, use a perceptually uniform colormap such as `viridis` for continuous data, and centre diverging data on zero. The default colour cycle is fine, but replacing it is a two-line change: ```python from cycler import cycler plt.rcParams['axes.prop_cycle'] = cycler( color=['#2563eb', '#f59e0b', '#10b981', '#ef4444', '#8b5cf6'] ) ``` For continuous data, choose a perceptually uniform colormap such as `viridis`, `magma` or `cividis`. Avoid `jet` and `rainbow`: Matplotlib's [colormap guide](https://matplotlib.org/stable/users/explain/colors/colormaps.html) explains that their lightness rises and falls, so they create visual boundaries that do not exist in the data and become unreadable in greyscale. For diverging data, centre the scale explicitly: ```python import matplotlib.colors as mcolors matrix = np.random.default_rng(1).uniform(-0.3, 1, size=(5, 5)) fig, ax = plt.subplots() norm = mcolors.TwoSlopeNorm(vmin=-1, vcenter=0, vmax=1) im = ax.imshow(matrix, cmap='RdBu_r', norm=norm) fig.colorbar(im, ax=ax, shrink=0.8) ``` Without `TwoSlopeNorm`, zero lands wherever the data happens to put it, and the colours mislead. In this example the values run from about -0.26 to 0.97, so a plain colour scale would put the neutral white at roughly 0.36 instead of at zero. ## Why is annotation the most valuable customisation? A chart that states its point beats a beautifully styled one that does not. ```python fig, ax = plt.subplots(figsize=(8, 4.5)) ax.plot(x, np.sin(x), color='#2563eb') peak_x = x[np.argmax(np.sin(x))] peak_y = np.sin(x).max() ax.annotate( 'first peak', xy=(peak_x, peak_y), xytext=(peak_x + 1.5, peak_y + 0.25), arrowprops={'arrowstyle': '->', 'color': '#475569'}, fontsize=10, ) ax.axhline(0, color='#94a3b8', linewidth=1, linestyle='--') ax.axvspan(6, 8, alpha=0.08, color='#2563eb') ax.text(7, -0.9, 'region of interest', ha='center', fontsize=9, color='#475569') ``` Here `np.argmax` finds the first peak at x ≈ 1.56, so the arrow points at the real maximum even if the data changes. Meanwhile, `axvspan` shades a range and `axhline` adds a reference line. ## Should you use a legend or label lines directly? Label lines directly when there are only a few of them. A legend forces the reader to look back and forth; a label where each line ends does not: ```python fig, ax = plt.subplots(figsize=(8, 4.5)) series = {'sin': np.sin(x), 'cos': np.cos(x)} colors = {'sin': '#2563eb', 'cos': '#f59e0b'} for name, y in series.items(): ax.plot(x, y, color=colors[name]) ax.text(x[-1] + 0.15, y[-1], name, color=colors[name], va='center', fontweight='bold') ax.set_xlim(0, 11.5) ax.spines[['top', 'right']].set_visible(False) ``` The extended `xlim` leaves room for the labels, and indexing `ax.spines` with a list hides both spines in one call. ## How do you save a Matplotlib figure without clipped labels? Pass `bbox_inches='tight'` to `savefig`, use `dpi=300` for PNG, and prefer PDF or SVG for documents: ```python fig.savefig('chart.png', dpi=300, bbox_inches='tight') fig.savefig('chart.pdf', bbox_inches='tight') # vector, scales cleanly fig.savefig('chart.svg', bbox_inches='tight') # vector, editable ``` `bbox_inches='tight'` prevents clipped axis labels, which is the most common export complaint. For anything going into a document or slide deck, prefer PDF or SVG, since they stay sharp at any size. As the [`savefig` reference](https://matplotlib.org/stable/api/_as_gen/matplotlib.pyplot.savefig.html) notes, `dpi` only affects raster formats such as PNG. ## Matplotlib styling quick reference | Goal | Setting or call | |---|---| | Default figure size | `'figure.figsize': (8, 4.5)` | | Sharp PNG exports | `'savefig.dpi': 300` | | No clipped labels | `'savefig.bbox': 'tight'` or `bbox_inches='tight'` | | Remove top/right spines | `'axes.spines.top': False`, `'axes.spines.right': False` | | Faint grid | `'axes.grid': True`, `'grid.alpha': 0.25` | | Seaborn-like look | `plt.style.use('seaborn-v0_8-whitegrid')` | | Temporary style | `with plt.style.context('dark_background'):` | | Custom palette | `cycler(color=[...])` on `axes.prop_cycle` | | Diverging colours centred on zero | `TwoSlopeNorm(vcenter=0)` | ## Common matplotlib styling mistakes - **Using `plt.style.use('seaborn')`.** The name was replaced by the `seaborn-v0_8-*` styles. - **Styling chart by chart.** Put shared settings in `rcParams` once. - **Choosing `jet` for continuous data.** Use `viridis` or `cividis` instead. - **Forgetting `bbox_inches='tight'`.** Labels get cut off in the saved file. - **Exporting slides as low-DPI PNGs.** Use PDF or SVG, or at least `dpi=300`. - **Mixing `plt.` and `ax.` calls in one figure.** Pick the object-oriented API and stick with it. ## Which three styling changes matter most? If you change only three things, use `fig, ax`, remove the top and right spines, and annotate the insight. Together, those cover more distance than every other tweak combined. ## Related reading For interactive charts instead, see the [Plotly tutorial, Part 1](https://www.1stepgrow.com/articles/plotly-tutorial) and [advanced Plotly charts in Part 2](https://www.1stepgrow.com/articles/advanced-plotly-charts). For the data layer underneath, the [NumPy tutorial series](https://www.1stepgrow.com/articles/numpy-tutorial) starts here. The fundamentals of figures and axes are covered in [Matplotlib figures, axes and subplots](https://www.1stepgrow.com/articles/matplotlib-figures-axes-subplots). ## Frequently asked questions ### What is the difference between plt.plot and ax.plot? plt.plot draws on whichever Axes pyplot currently considers active, which is convenient in a notebook but fragile in scripts with several figures. ax.plot targets one specific Axes object explicitly. Matplotlib's own documentation notes that complicated plots often end up simpler with the explicit Axes interface, so use it for anything beyond a throwaway chart. ### How do I make matplotlib charts look less dated? Three changes cover most of it. First, pick a modern style with plt.style.use, such as seaborn-v0_8-whitegrid. Second, remove the top and right spines through rcParams. Third, lighten the grid with a low alpha. After that, set a sensible default figure size and DPI once so every chart in the notebook inherits them. ### How do I export a matplotlib chart for print? Call savefig with dpi=300 for a raster PNG, or save as PDF or SVG for vector output that stays sharp at any size; the dpi setting does not affect vector formats. Always pass bbox_inches='tight' so axis labels and titles are not clipped at the edge of the saved file, which is the most common export complaint. ### Should I use matplotlib or seaborn? Seaborn is built on matplotlib and gives better defaults with less code for statistical charts. Use seaborn to draw and matplotlib to customise; they compose well because seaborn's axes-level functions, such as histplot and scatterplot, accept and return a matplotlib Axes. Everything in this guide, from rcParams to annotation, therefore still applies to a seaborn chart. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/matplotlib-styling-customization/). Cite with the title, "1stepGrow" and a link._ --- # Python Classes and Objects: Object-Oriented Programming, Part 1 > Python classes and objects explained with tested code: __init__, self, dunder methods, properties and dataclasses, plus the shared-list bug to learn first. - **Author:** Althaf Ashraf — AI Systems Engineer, Tata Consultancy Services (https://www.1stepgrow.com/authors/althaf-ashraf/) - **Published:** Jul 29, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/python-classes-objects/ ## Key takeaways - A class bundles data with the functions that operate on it, so reach for one when those two keep travelling together. - self is just the instance, passed automatically when you call a method; the name is a convention, not a keyword. - Mutable class attributes are shared across all instances, which is the classic OOP bug in Python. - Comparison and arithmetic dunder methods should return NotImplemented for types they do not handle. - Not everything needs a class, and a plain function or a dataclass is often the better answer. Python classes and objects are how you bundle data with the functions that work on that data. A class is the blueprint; an object, or instance, is one thing built from it with its own state. Use a class when data and behaviour keep travelling together; otherwise a plain function is simpler. Classes also bring bugs that plain functions do not. Define a list in the class body and every instance shares it, so one portfolio's stock appears in another's. This part is for Python programmers who are comfortable with functions and want the object-oriented features they will actually use: `__init__` and `self`, class versus instance attributes, dunder methods, properties, and when a dataclass or a function beats a class. [Part 2 covers inheritance and composition](https://www.1stepgrow.com/articles/python-inheritance-composition). ## How do you define Python classes and objects? Write `class Name:` with an `__init__` method that sets attributes on `self`, add methods, then call the class like a function to create an object: ```python class BankAccount: def __init__(self, owner, balance=0): self.owner = owner self.balance = balance def deposit(self, amount): if amount <= 0: raise ValueError("Deposit must be positive") self.balance += amount return self.balance def withdraw(self, amount): if amount > self.balance: raise ValueError("Insufficient funds") self.balance -= amount return self.balance ``` Using it: ```python acct = BankAccount("Ananya", 1000) print(acct.deposit(500)) # 1500 print(acct.withdraw(200)) # 1300 print(acct.owner) # Ananya ``` Three things are happening. - **`__init__`** runs when you create an instance. It is not a constructor in the C++ sense, because the object already exists. According to the [data model reference](https://docs.python.org/3/reference/datamodel.html#object.__init__), it is called after the instance has been created, so it is an initialiser that sets up state. - **`self`** is the instance. In fact, `acct.deposit(500)` is really `BankAccount.deposit(acct, 500)`, because Python passes the instance as the first argument for you. That is the entire mechanism. - **Attributes assigned on `self`** belong to that instance. As a result, two accounts have independent balances. ## Why are mutable class attributes a trap? Because a list or dict defined in the class body is a single object shared by every instance, so a change through one instance shows up in all of them. This is the bug worth learning before you hit it. ```python class Portfolio: holdings = [] # class attribute - shared by ALL instances def add(self, stock): self.holdings.append(stock) a = Portfolio() b = Portfolio() a.add("INFY") print(b.holdings) # ['INFY'] <- b sees a's stock ``` `holdings` was defined in the class body, so there is exactly one list shared by every instance. Consequently, appending through one instance is visible from all of them. The [Python tutorial on class and instance variables](https://docs.python.org/3/tutorial/classes.html#class-and-instance-variables) uses almost the same example with a list of dog tricks. The fix is to create the mutable object per instance, in `__init__`: ```python class Portfolio: def __init__(self): self.holdings = [] # a new list for each instance def add(self, stock): self.holdings.append(stock) ``` Class attributes are still genuinely useful for constants shared across instances: ```python class Circle: PI = 3.14159 # fine - immutable, genuinely shared def __init__(self, radius): self.radius = radius def area(self): return Circle.PI * self.radius ** 2 print(Circle(2).area()) # 12.56636 ``` In real code you would use `math.pi`, but the principle holds. The rule: **immutable shared values as class attributes, mutable state on self.** ## What are dunder methods? Dunder methods are special methods named with double underscores, such as `__repr__`, `__eq__` and `__add__`. They let your objects work with built-in syntax such as `+`, `==` and `print()`. ```python class Money: def __init__(self, amount, currency="INR"): self.amount = amount self.currency = currency def __repr__(self): return f"Money({self.amount}, {self.currency!r})" def __str__(self): return f"{self.currency} {self.amount:,.2f}" def __add__(self, other): if not isinstance(other, Money): return NotImplemented if self.currency != other.currency: raise ValueError("Currency mismatch") return Money(self.amount + other.amount, self.currency) def __eq__(self, other): if not isinstance(other, Money): return NotImplemented return (self.amount, self.currency) == (other.amount, other.currency) ``` ```python a = Money(1500) b = Money(2500) print(a + b) # INR 4,000.00 print(repr(a)) # Money(1500, 'INR') print(a == Money(1500)) # True print(a == 1500) # False print([a, b]) # [Money(1500, 'INR'), Money(2500, 'INR')] a + 5 # TypeError: unsupported operand type(s) for +: 'Money' and 'int' ``` The `isinstance` checks matter. Without them, `a == 1500` crashes with an `AttributeError` because an integer has no `amount`. Returning `NotImplemented` tells Python to try the other operand and, failing that, to fall back to a sensible default or a clear `TypeError`. `__repr__` is for developers and should ideally be unambiguous; `__str__` is for users. Define `__repr__` at minimum, since it is what shows up in debuggers, logs and lists, and the default `<__main__.Money object at 0x...>` tells you nothing. One side effect to know: a class that defines `__eq__` without `__hash__` becomes unhashable, so `{a}` raises `TypeError`. That is deliberate. If you need `Money` in sets or as dictionary keys, define `__hash__` from the same fields and treat the object as immutable. ## How do properties work? Use a property when you want an attribute that is computed, or validated on assignment, without changing the calling code: ```python class Temperature: def __init__(self, celsius=0): self.celsius = celsius # goes through the setter @property def celsius(self): return self._celsius @celsius.setter def celsius(self, value): if value < -273.15: raise ValueError("Below absolute zero") self._celsius = value @property def fahrenheit(self): return self._celsius * 9 / 5 + 32 ``` ```python t = Temperature(25) print(t.fahrenheit) # 77.0 - looks like an attribute, runs a method t.celsius = 30 # validated t.celsius = -300 # ValueError: Below absolute zero Temperature(-500) # ValueError: Below absolute zero t.fahrenheit = 100 # AttributeError - read-only property ``` Notice that `__init__` assigns `self.celsius`, not `self._celsius`. Writing to the private name directly would skip the setter, so `Temperature(-500)` would quietly succeed. The leading underscore on `_celsius` is a convention meaning "internal, do not touch from outside". Python does not enforce it. A double underscore, such as `__celsius`, triggers name mangling to `_Temperature__celsius`, which avoids clashes in subclasses but is not real privacy either. ## When should you not write a class? Plenty of code is worse as a class. If your class has an `__init__` and exactly one method, it wanted to be a function. ```python # over-engineered class DataCleaner: def __init__(self, df): self.df = df def clean(self): return self.df.dropna().drop_duplicates() # better def clean(df): return df.dropna().drop_duplicates() ``` For plain data containers, a dataclass removes the boilerplate: ```python from dataclasses import dataclass, field @dataclass class Point: x: float y: float p = Point(1.0, 2.0) print(p) # Point(x=1.0, y=2.0) print(p == Point(1.0, 2.0)) # True - __repr__ and __eq__ for free @dataclass class Basket: items: list = field(default_factory=list) # not items: list = [] ``` The [dataclasses documentation](https://docs.python.org/3/library/dataclasses.html#mutable-default-values) explains why: a bare `[]` default raises `ValueError`, which protects you from the shared-list trap above. ## Quick reference: class building blocks | Feature | What it is for | Example | |---|---|---| | `__init__` | Set up instance state | `self.balance = balance` | | Instance attribute | Data owned by one object | `self.holdings = []` | | Class attribute | Constant shared by all instances | `PI = 3.14159` | | `__repr__` / `__str__` | Developer and user text | `Money(1500, 'INR')` / `INR 1,500.00` | | `__eq__`, `__add__` | Make `==` and `+` work | Return `NotImplemented` for other types | | `@property` | Computed or validated attribute | `t.fahrenheit` | | `@dataclass` | Data container without boilerplate | `Point(x=1.0, y=2.0)` | ## Common mistakes with classes - **Forgetting `self` in a method definition.** `def deposit(amount)` fails when called with `TypeError: BankAccount.deposit() takes 1 positional argument but 2 were given`, because the instance is passed as well. - **Mutable class attributes.** Put lists and dictionaries on `self` in `__init__`. - **Bypassing your own validation.** Assign through the property in `__init__`, as `Temperature` does. - **Returning a value from `__init__`.** It must return `None`, otherwise Python raises `TypeError`. - **Classes with one method.** Write a function instead. ## What comes after classes and objects? [Part 2 of this series](https://www.1stepgrow.com/articles/python-inheritance-composition) covers inheritance, composition, class and static methods, and abstract base classes, plus why "prefer composition over inheritance" is advice worth taking seriously. ## Related reading [Inheritance and composition in Python (OOP Part 2)](https://www.1stepgrow.com/articles/python-inheritance-composition) continues the series. For the fundamentals underneath, see [Python's building blocks](https://www.1stepgrow.com/articles/python-basics). Custom exception classes are a common first use of classes, covered in [exception handling in Python](https://www.1stepgrow.com/articles/python-exception-handling). ## Frequently asked questions ### When should I write a class instead of a function? Write a class when you have state that several functions need to share and change over time. If you find yourself passing the same three arguments into every function in a module, those arguments probably want to be a class. If your code is a transformation with no persistent state, keep it a function. ### What is self and why do I have to write it? self is the instance the method was called on. Python passes it automatically when you call obj.method(), and the explicit self in the definition makes that binding visible. The Python tutorial calls the name nothing more than a convention, but every tool and reader expects it, so never rename it. ### What is the difference between a class attribute and an instance attribute? A class attribute is defined in the class body and shared by every instance. An instance attribute is assigned on self, usually in __init__, and belongs to that object alone. Immutable shared constants work well as class attributes, but mutable ones such as lists cause surprising shared-state bugs. ### Do data scientists need OOP? You need to read it, because pandas, scikit-learn and most libraries you use are built with classes. You will write classes occasionally, such as a custom transformer or a small data container. Most analysis code, however, is clearer as a set of functions, so do not force a class where none is needed. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/python-classes-objects/). Cite with the title, "1stepGrow" and a link._ --- # Inheritance and Composition in Python: OOP in Python, Part 2 > Inheritance and composition in Python with tested code: when to use each, how super() really works, duck typing, and why deep class hierarchies backfire. - **Author:** Althaf Ashraf — AI Systems Engineer, Tata Consultancy Services (https://www.1stepgrow.com/authors/althaf-ashraf/) - **Published:** Jul 28, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/python-inheritance-composition/ ## Key takeaways - Inheritance models an 'is a' relationship and composition models 'has a'; most real relationships are 'has a'. - super() follows the method resolution order, so always use it rather than naming the parent class directly. - Duck typing means you often do not need inheritance at all, because a shared method name is enough. - Abstract base classes make a missing method fail at instantiation rather than deep inside a call stack. - Two levels of inheritance is usually plenty, because deep hierarchies become impossible to reason about. Inheritance and composition in Python are the two ways classes relate to each other. Inheritance makes one class a specialised version of another, while composition builds a class from other objects it holds. Prefer composition by default and use inheritance for genuine "is a" relationships kept shallow. The syntax is the easy part. The design mistakes cost more: naming the parent class instead of calling `super()` can silently skip a class under multiple inheritance, and a five-level hierarchy makes it hard to find which class a method comes from. This part is for readers who know [Python classes and objects from Part 1](https://www.1stepgrow.com/articles/python-classes-objects). It covers `super()` and the MRO, composition, duck typing, class and static methods and abstract base classes, with a table for choosing between them. ## How does inheritance work in Python? A subclass gets the parent's attributes and methods, and can add or override them: ```python class Employee: def __init__(self, name, salary): self.name = name self.salary = salary def describe(self): return f"{self.name} earns {self.salary:,}" def annual_cost(self): return self.salary * 12 class Manager(Employee): def __init__(self, name, salary, reports): super().__init__(name, salary) # run the parent's setup self.reports = reports def describe(self): # override base = super().describe() return f"{base} and manages {len(self.reports)} people" ``` ```python m = Manager("Rekha", 180_000, ["Amit", "Sara"]) print(m.describe()) # Rekha earns 180,000 and manages 2 people print(m.annual_cost()) # 2160000 - inherited unchanged ``` Two things to note. First, `super().__init__(...)` runs the parent initialiser rather than duplicating it. Second, `super().describe()` inside the override lets you extend behaviour instead of replacing it. Forget the `super().__init__` call and the parent's attributes never get set. Accessing `m.name` would then raise `AttributeError: 'Manager' object has no attribute 'name'`. ## Why use super() instead of naming the parent? Always use `super()` rather than `Employee.__init__(self, ...)`. The [super() documentation](https://docs.python.org/3/builtins/functions.html#super) describes it as returning a proxy that delegates to "a parent or sibling class", following the method resolution order (MRO). Under multiple inheritance, the next class is not necessarily the one you would have named: ```python class Base: def greet(self): return "Base" class Logged(Base): def greet(self): return "Logged > " + super().greet() class Cached(Base): def greet(self): return "Cached > " + super().greet() class Service(Logged, Cached): def greet(self): return "Service > " + super().greet() print(Service().greet()) # Service > Logged > Cached > Base print([k.__name__ for k in Service.__mro__]) # ['Service', 'Logged', 'Cached', 'Base', 'object'] ``` Inside `Logged`, `super()` points to `Cached`, a sibling, not to `Base`. Consequently every class runs exactly once. Had `Logged` called `Base.greet(self)` directly, `Cached` would have been skipped. The [tutorial section on multiple inheritance](https://docs.python.org/3/tutorial/classes.html#multiple-inheritance) explains how the order is computed. ## Why does composition usually win over inheritance? Inheritance couples classes tightly. The subclass depends on the parent's internals, and changing the parent can break children in non-obvious ways. This is known as the "fragile base class" problem. Composition holds a reference instead: ```python class Engine: def __init__(self, horsepower): self.horsepower = horsepower def start(self): return "Engine started" class ElectricMotor: def __init__(self, kilowatts): self.kilowatts = kilowatts def start(self): return "Motor humming" class Car: def __init__(self, model, power_unit): self.model = model self.power_unit = power_unit # has-a, not is-a def start(self): return f"{self.model}: {self.power_unit.start()}" ``` ```python print(Car("Nexon", Engine(120)).start()) # Nexon: Engine started print(Car("Nexon EV", ElectricMotor(95)).start()) # Nexon EV: Motor humming ``` The power unit is swappable at runtime, and `Car` does not depend on either class's internals. Testing is also easy, because you can pass a fake object with a `start` method. The heuristic: **if you would say "is a", consider inheritance. If you would say "has a", use composition.** Most real relationships are "has a". ## Inheritance and composition in Python compared | Question | Inheritance | Composition | Duck typing | |---|---|---|---| | Relationship | "is a" | "has a" | "behaves like" | | Coupling | Tight: subclass sees parent internals | Loose: talks through methods | None: only the method name matters | | Swap behaviour at runtime | No | Yes | Yes | | Code reuse | Automatic | Explicit delegation | None | | Typical Python use | Custom exceptions, framework base classes | Services, pipelines, models with parts | Exporters, file-like objects, plugins | ## How does duck typing remove the need for inheritance? Python does not require a shared base class for polymorphism. If two objects have the same method, they are interchangeable: ```python import json class CSVExporter: def export(self, rows): return "\n".join(",".join(map(str, r)) for r in rows) class JSONExporter: def export(self, rows): return json.dumps(rows) def save(exporter, rows): return exporter.export(rows) # neither class inherits from anything rows = [[1, "a"], [2, "b"]] print(save(CSVExporter(), rows)) # 1,a then 2,b on the next line print(save(JSONExporter(), rows)) # [[1, "a"], [2, "b"]] ``` In a statically typed language this would need an interface. In Python, having the method is enough, which removes a large share of the inheritance you might otherwise write. If you want a type checker to verify the shape, [typing.Protocol](https://docs.python.org/3/library/typing.html#typing.Protocol) describes it without forcing the classes to inherit anything. ## What is the difference between a class method and a static method? A `classmethod` receives the class as `cls` and suits alternative constructors; a `staticmethod` receives nothing and is a plain function kept in the class: ```python class Date: def __init__(self, day, month, year): self.day, self.month, self.year = day, month, year @classmethod def from_string(cls, text): d, m, y = map(int, text.split("-")) return cls(d, m, y) # cls, so subclasses get their own type @staticmethod def is_leap(year): return year % 4 == 0 and (year % 100 != 0 or year % 400 == 0) ``` ```python d = Date.from_string("15-08-2026") print(d.day, d.month, d.year) # 15 8 2026 print(Date.is_leap(2028)) # True print(Date.is_leap(2100)) # False class IndianDate(Date): pass print(type(IndianDate.from_string("15-08-2026")).__name__) # IndianDate ``` A `classmethod` receives the class and is the idiomatic way to write alternative constructors. Using `cls(...)` rather than `Date(...)` means a subclass calling `from_string` gets an instance of itself, as the last line shows. By contrast, a `staticmethod` receives nothing automatically. It is a plain function that lives in the class for organisational reasons. ## When are abstract base classes worth it? Use one when you want to define an interface and fail loudly if it is not implemented: ```python from abc import ABC, abstractmethod class Model(ABC): @abstractmethod def fit(self, X, y): ... @abstractmethod def predict(self, X): ... def score(self, X, y): # concrete, shared by all subclasses preds = self.predict(X) return sum(p == t for p, t in zip(preds, y)) / len(y) class Baseline(Model): def fit(self, X, y): self.majority = max(set(y), key=list(y).count) return self def predict(self, X): return [self.majority] * len(X) ``` ```python X = [[0], [1], [2], [3]] y = ["spam", "ham", "ham", "ham"] print(Baseline().fit(X, y).score(X, y)) # 0.75 Model() # TypeError: Can't instantiate abstract class Model without an # implementation for abstract methods 'fit', 'predict' ``` As the [abc module documentation](https://docs.python.org/3/library/abc.html#abc.abstractmethod) states, a class with abstract methods cannot be instantiated until all of them are overridden. A subclass missing `predict` therefore fails at instantiation rather than deep inside a call stack later, which is the whole point. The exact wording of the error message varies slightly between Python versions. ## Common mistakes with inheritance - **Forgetting `super().__init__()`.** The parent's attributes are never set. - **Hardcoding the parent class.** `Employee.__init__(self, ...)` breaks cooperative multiple inheritance. - **Inheriting to reuse one helper method.** Move the helper into a function or a component instead. - **Returning `Date(...)` from a classmethod.** Subclasses then get the wrong type; return `cls(...)`. - **Deep hierarchies.** Five levels of subclasses make it hard to find which class implements a method. ## How deep should an inheritance hierarchy go? Two levels is usually plenty. The most common OOP failure in real codebases is not wrong syntax. Instead, it is a five-level inheritance chain where finding which class actually implements a method requires reading all five. If you need more, that is a strong signal that some of those relationships are "has a" and want composition instead. Practical guidance: prefer functions to classes, prefer composition to inheritance, prefer shallow to deep, and only add abstraction when the duplication actually hurts. ## Related reading [Python classes and objects (OOP Part 1)](https://www.1stepgrow.com/articles/python-classes-objects) covers classes, instances and dunder methods. Custom error hierarchies are a good place to practise inheritance, as shown in [exception handling in Python](https://www.1stepgrow.com/articles/python-exception-handling). For functional patterns, see [lambda functions](https://www.1stepgrow.com/articles/python-lambda-functions) and [iterators and generators](https://www.1stepgrow.com/articles/python-iterators-generators). ## Frequently asked questions ### What is the difference between inheritance and composition? Inheritance says a Manager is an Employee and gets its behaviour automatically. Composition says a Car has an Engine and delegates to it. Composition is usually more flexible, because you can swap the component at runtime and the two classes do not depend on each other's internals, which also makes testing with fakes easy. ### What does super() actually do? super() returns a proxy object that delegates method calls to the next class in the method resolution order. Under multiple inheritance that next class can be a sibling rather than the direct parent. That is precisely why you should use super() instead of hardcoding ParentClass.method(self), which skips classes and breaks cooperative designs. ### What is the difference between a class method and a static method? A classmethod receives the class itself as its first argument, usually named cls, and is typically used for alternative constructors such as from_string. A staticmethod receives nothing automatically. It is really just a plain function grouped inside the class for organisation, and it cannot see the class or instance unless you pass them. ### Do I need abstract base classes? Rarely in small codebases. They earn their place when you are defining an interface that several people or plugins will implement, and you want a clear TypeError at instantiation rather than an AttributeError deep in a call stack. For lighter-weight checking with type checkers, typing.Protocol describes the same interface without inheritance. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/python-inheritance-composition/). Cite with the title, "1stepGrow" and a link._ --- # Python Tuples: Unpacking, Dict Keys and the Trailing-Comma Trap > How to use Python tuples well: creating them, unpacking, using them as dict keys, named tuples, and the one-element and += gotchas that catch everybody. - **Author:** Ayushi Kulshreshta — AI Engineer (https://www.1stepgrow.com/authors/ayushi-kulshreshta/) - **Published:** Jul 27, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/python-tuples/ ## Key takeaways - Tuples are immutable, which makes them hashable — so they can be dictionary keys and set members where lists cannot. - A one-element tuple needs the trailing comma: (5,) not (5). - Immutability is shallow — a tuple containing a list still lets that list be modified, and such a tuple is not hashable. - Unpacking, including the starred form, is the tuple feature you will use most. - Use namedtuple or typing.NamedTuple when positional access starts hurting readability. Python tuples are ordered, immutable sequences, written with commas and usually parentheses: `(3, 4)`. Their immutability is the point, not a limitation: it is what lets a tuple act as a dictionary key or set member, which a list never can. The catch is that `(5)` is not a tuple at all, and a tuple holding a list is not truly immutable. This is the hands-on guide to tuples themselves, for readers who already know lists. You will learn to create and unpack them, key dictionaries on them, switch to named tuples when positions get unreadable, and avoid the five mistakes listed near the end. If you are only deciding between the two sequence types, [lists vs tuples in Python](https://www.1stepgrow.com/articles/lists-vs-tuples-python) is the quicker read. ## How do you create Python tuples? ```python point = (3, 4) rgb = (255, 128, 0) empty = () mixed = (1, "two", 3.0, [4, 5]) # parentheses are optional coords = 3, 4 # from any iterable tuple("abc") # ('a', 'b', 'c') tuple([1, 2]) # (1, 2) ``` The gotcha that catches everyone: ```python not_a_tuple = (5) # this is the integer 5 actual_tuple = (5,) # this is a tuple - note the comma type(not_a_tuple) # type(actual_tuple) # ``` **The comma creates the tuple, not the parentheses.** The official [tutorial on tuples and sequences](https://docs.python.org/3/tutorial/datastructures.html#tuples-and-sequences) describes the trailing comma as ugly but effective. It is worth saying twice, because the resulting bugs are quiet: a function that returns `(value)` instead of `(value,)` returns a different type than its caller expects. ## How does tuple unpacking work? Unpacking assigns each element of a tuple to its own name in one statement, as in `x, y = point`. It is the tuple feature you will use most, often without thinking of it as tuples: ```python point = (3, 4) x, y = point # swapping, no temporary variable a, b = 1, 2 a, b = b, a # functions returning multiple values return a tuple def min_max(nums): return min(nums), max(nums) low, high = min_max([3, 1, 4, 1, 5]) # low = 1, high = 5 ``` In fact, you already unpack tuples every time you loop over `enumerate()` or `dict.items()`, because both yield pairs: ```python list(enumerate("xy")) # [(0, 'x'), (1, 'y')] list({"a": 1}.items()) # [('a', 1)] ``` Extended unpacking with a star collects the remainder into a list: ```python first, *rest = (1, 2, 3, 4, 5) # first = 1, rest = [2, 3, 4, 5] first, *middle, last = (1, 2, 3, 4, 5) # first = 1, middle = [2, 3, 4], last = 5 _, important, _ = ("skip", "keep", "skip") a, b = (1, 2), (3,) (*a, *b, 4) # (1, 2, 3, 4) - star-unpacking into a new tuple ``` Nested unpacking works too, which is handy when iterating structured data: ```python records = [("Ananya", (85, 92)), ("Rohit", (78, 88))] for name, (maths, science) in records: print(f"{name}: {maths + science}") ``` ```text Ananya: 177 Rohit: 166 ``` ## Why does tuple immutability matter? **Tuples are hashable, so they can be dictionary keys and set members.** Lists cannot. The [glossary definition of hashable](https://docs.python.org/3/glossary.html#term-hashable) adds the condition: an immutable container is hashable only if its elements are hashable too. ```python locations = { (19.0760, 72.8777): "Mumbai", (28.6139, 77.2090): "Delhi", } print(locations[(19.0760, 72.8777)]) # Mumbai # with a list as the key this raises TypeError (unhashable type: 'list') ``` This is genuinely useful for keying on compound values, such as a (user_id, date) pair, a grid coordinate or a category combination. ```python from collections import defaultdict sales = defaultdict(int) sales[("North", "Q1")] += 1200 sales[("North", "Q2")] += 1450 ``` As a worked example, `Counter` can count combinations directly when each combination is a tuple: ```python from collections import Counter orders = [("Pune", "Laptop"), ("Delhi", "Phone"), ("Pune", "Laptop"), ("Pune", "Phone")] Counter(orders).most_common(2) # [(('Pune', 'Laptop'), 2), (('Delhi', 'Phone'), 1)] ``` **Tuples also signal intent.** A reader seeing a tuple knows the contents are not expected to change. That is documentation the language enforces. ## Can the contents of a tuple change? Yes, if an element is itself mutable. A tuple's immutability is shallow: it fixes which objects the tuple holds, not the state of those objects. ```python t = (1, 2, [3, 4]) t[2].append(5) print(t) # (1, 2, [3, 4, 5]) ``` The tuple itself is unchanged, because it still references the same three objects. But the list inside is mutable, and nothing stops you modifying it. The practical consequence is that such a tuple is not hashable either: ```python hash((1, 2, [3, 4])) # TypeError: unhashable type: 'list' hash((1, 2, (3, 4))) # fine - returns an int ``` The strangest version of this is `+=` on a list inside a tuple. It raises an error **and** changes the list, a case the Python FAQ explains under [why `a_tuple[i] += ['item']` raises an exception when the addition works](https://docs.python.org/3/faq/programming.html#why-does-a-tuple-i-item-raise-an-exception-when-the-addition-works): ```python t = (1, [2]) t[1] += [3] # TypeError: 'tuple' object does not support item assignment print(t) # (1, [2, 3]) - the list was extended anyway ``` If you need a genuinely immutable nested structure, therefore, use tuples all the way down. ## When should you switch to a named tuple? Switch as soon as positional access stops explaining itself: `person[2]` tells a reader nothing, while `person.city` does. A named tuple adds field names without giving up any tuple behaviour. ```python from collections import namedtuple Person = namedtuple("Person", ["name", "age", "city"]) p = Person("Ananya", 28, "Pune") print(p.name) # Ananya - readable print(p[0]) # Ananya - still works positionally print(p) # Person(name='Ananya', age=28, city='Pune') ``` They remain tuples, so unpacking and hashing still work. Moreover, the [`collections.namedtuple` documentation](https://docs.python.org/3/library/collections.html#collections.namedtuple) notes that instances have no per-instance dictionary and need no more memory than regular tuples. The typed version, [`typing.NamedTuple`](https://docs.python.org/3/library/typing.html#typing.NamedTuple), reads better in modern code: ```python from typing import NamedTuple class Point(NamedTuple): x: float y: float def distance_from_origin(self) -> float: return (self.x ** 2 + self.y ** 2) ** 0.5 p = Point(3, 4) p.distance_from_origin() # 5.0 ``` Useful methods: ```python p._asdict() # {'x': 3, 'y': 4} p._replace(x=10) # Point(x=10, y=4) - returns a new instance ``` If you need mutability or more behaviour, a `@dataclass` is the natural next step — see our [OOP guide](https://www.1stepgrow.com/articles/python-classes-objects). ## Tuple methods and operations Tuples have only two methods, because there is nothing to modify: ```python t = (1, 2, 2, 3, 2) t.count(2) # 3 t.index(2) # 1 - first occurrence ``` Everything else is standard sequence behaviour, summarised here: | Operation | Example | Result | |---|---|---| | Length | `len(t)` | `5` | | Concatenation | `t + (4, 5)` | a **new** tuple | | Repetition | `(0,) * 3` | `(0, 0, 0)` | | Membership | `3 in t` | `True` | | Slicing | `t[1:3]` | `(2, 2)` | | Sorting | `sorted(t)` | a **list**: `[1, 2, 2, 2, 3]` | | Comparison | `(1, 2, 3) < (1, 3)` | `True` - compared item by item | Note that `t + (4, 5)` builds a new tuple. Consequently, repeatedly concatenating in a loop is O(n²) — use a list to accumulate and convert at the end. ## Common mistakes with Python tuples - **Forgetting the trailing comma** in a one-element tuple. - **Assuming a tuple is deeply immutable** when it holds a list or dict. - **Using a tuple containing a list as a dict key**, which raises `TypeError`. - **Growing a tuple in a loop** with `+`, instead of appending to a list and calling `tuple()` once. - **Relying on positions like `row[7]`** when a named tuple would say what the field is. ## When should you use a tuple? **Use a tuple** for fixed-size heterogeneous records, multiple return values, dictionary keys and set members, and anywhere you want to signal that something does not change. **Use a list** for homogeneous collections that grow, shrink or get sorted in place. The [lists vs tuples comparison](https://www.1stepgrow.com/articles/lists-vs-tuples-python) goes deeper on that decision, including memory and the mutable default argument bug. ## Related reading [Lists and tuples compared](https://www.1stepgrow.com/articles/lists-vs-tuples-python) covers when to use each. [Python basics](https://www.1stepgrow.com/articles/python-basics) covers the wider set of core types. ## Frequently asked questions ### What is the difference between a tuple and a list? Lists are mutable and unhashable; tuples are immutable and hashable, provided their contents are hashable. That single difference drives everything else: tuples can be dictionary keys, they are slightly smaller, and they signal to a reader that the contents are not meant to change. Our lists vs tuples comparison covers when to choose each. ### Why does (5) not create a tuple? Because the parentheses are just grouping, so (5) is the integer 5. The comma creates the tuple, not the brackets. Write (5,) instead, or even 5, on its own, which is also a valid one-element tuple. The empty tuple () is the only tuple that needs no comma at all. ### Can I modify a tuple in Python? No, you cannot rebind its elements, so t[0] = 1 raises TypeError. However, immutability is shallow: if a tuple contains a mutable object such as a list, that object can still change in place. To get a modified tuple, build a new one, for example with concatenation or a named tuple's _replace method. ### Are tuples faster than lists? Slightly, for creation, and they use a little less memory because they never over-allocate. Iteration speed is essentially the same. The difference rarely matters in real programs, so choose based on whether the data should be immutable or needs to act as a dictionary key, not on performance. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/python-tuples/). Cite with the title, "1stepGrow" and a link._ --- # AWS vs Azure Certification in 2026: Which Path Pays Off for You? > AWS vs Azure certification in 2026: exam fees, prerequisites, renewal rules and three-year costs compared, plus a 20-job-posting test to pick the right cloud. - **Author:** Sivaranjani S — Cloud Operations Engineer, Zoho (https://www.1stepgrow.com/authors/sivaranjani-s/) - **Published:** Jul 26, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Cloud Computing · **Format:** Comparison · **Read time:** 8 min - **Canonical URL:** https://www.1stepgrow.com/articles/aws-vs-azure-certification-path/ ## Key takeaways - AWS has more job listings globally. Azure dominates large enterprises, BFSI and government — including in India. - Azure's official study material is free and better structured. AWS has a larger third-party ecosystem. - The certification is worth far less than the environment you use daily. Pick the cloud your employers run. - Total cost over three years is closer than the exam fees suggest, once labs and renewals are counted. The AWS vs Azure certification decision comes down to one question: which cloud do the employers you want actually run? Both tracks are credible and take a similar number of evenings. AWS certifications last three years, while Azure ones renew free every year, and that difference matters less than hands-on experience. Pick one and start. The gap between the two tracks is smaller than the gap between "certified with hands-on experience" and "certified without it". Timing matters too: both vendors retired or refreshed exams in 2026, so study material bought last year may point at an exam you can no longer sit. This comparison is for engineers and career switchers choosing a first cloud credential, or adding a second. It covers every level side by side, what renewal costs over three years, the 2026 exam changes, and a quick way to settle the choice using job postings in your own city. | Feature | Learnbay Cloud & DevOps Master Program | AWS certification track | Azure certification track | | --- | --- | --- | --- | | What you actually get | Taught programme + 12 deployed projects | An exam result | An exam result | | Cloud coverage | Multi-cloud, vendor-neutral | AWS only | Azure only | | Global job listing volume | Maps to both tracks | Highest | Second, but close in enterprise | | Enterprise & BFSI penetration | Both, via the exam you choose | Strong | Dominant | | Structured teaching | Live cohort with mentor review | Self-study | Self-study (Microsoft Learn) | | Hands-on project work | 12+ real domain projects | Labs you arrange yourself | Labs you arrange yourself | | Prerequisite chain | None | None enforced | AZ-104 required for the AZ-305 Expert certification | | Renewal effort | No expiry on the programme itself | Every 3 years: pass the current or a higher exam | Every year: free online assessment | | Interview & placement support | Mock interviews and job calls | None | None | | Cost | On request — the most expensive of the three | $150-$300 per exam | $165 per exam | **Verdict — Learnbay Cloud & DevOps Master Program:** The two vendor tracks are not really competitors to each other — you pick whichever cloud your employers run, and this article's core advice on that stands. They are also not a complete answer, because both certify recall rather than capability, and the thing that actually gets you hired is deployed architecture you can talk through. Learnbay's programme sits at the top here because it produces that portfolio and prepares you for whichever exam you then sit. If budget is the binding constraint, take the vendor exam for your employer's cloud and build the portfolio yourself — that path is slower and considerably cheaper. ## AWS vs Azure certification paths, level by level Both vendors offer a foundational, an associate and a senior tier, but the senior tiers work differently. The table below uses each vendor's official pages as of September 2026, with US exam prices before tax. | Level | AWS | Azure | |---|---|---| | Foundational | [Cloud Practitioner](https://aws.amazon.com/certification/certified-cloud-practitioner/): $100, 90 minutes | [Azure Fundamentals (AZ-900)](https://learn.microsoft.com/en-us/credentials/certifications/azure-fundamentals/): 45 minutes, priced by country | | Associate, infrastructure | [Solutions Architect – Associate](https://aws.amazon.com/certification/certified-solutions-architect-associate/): $150, 130 minutes | [Azure Administrator (AZ-104)](https://learn.microsoft.com/en-us/credentials/certifications/azure-administrator/): $165, 100 minutes | | Senior architect | Solutions Architect – Professional: $300, 180 minutes, no prerequisite | [Solutions Architect Expert (AZ-305)](https://learn.microsoft.com/en-us/credentials/certifications/azure-solutions-architect/): $165, and you must hold AZ-104 | | Senior DevOps | DevOps Engineer – Professional: $300, 180 minutes | [DevOps Engineer Expert (AZ-400)](https://learn.microsoft.com/en-us/credentials/certifications/devops-engineer/): requires the Azure Administrator or Azure Developer associate first | | Validity | 3 years | 1 year for role-based certifications; fundamentals do not expire | | Renewal | Pass the current or a higher exam, or pay for a one-year extension through Skill Builder | Free, unproctored online assessment | Two structural differences stand out. First, AWS lets you sit a professional exam without holding an associate one, while Microsoft enforces a prerequisite for its expert certifications. Second, AWS charges more per senior exam but asks you to renew only every three years. ## How does renewal change the real cost? Renewal barely changes the money but changes the effort a lot. It is where the two tracks differ most, so count it before you compare exam fees. - **AWS** certifications are valid for three years. To [recertify](https://aws.amazon.com/certification/recertification/), you pass the latest version of the exam or a higher-level one; passing Solutions Architect – Professional automatically recertifies the Associate. Holders also get a 50% discount on their next AWS exam. - **Azure** role-based certifications expire after a year, but [renewal is free](https://learn.microsoft.com/en-us/credentials/certifications/renew-your-microsoft-certification): you pass a short, open-book assessment at any point in the six months before expiry. Here is what three years looks like on each track, using only the official US prices: | Three-year path | AWS | Azure | |---|---|---| | Associate only | $150, then renew at year three | $165, plus two free renewals | | Associate, then architect | $150 + $300 = $450 at list price, or $300 with the 50% holder discount on the second exam | $165 + $165 = $330, plus free renewals | | Architect only | $300, since AWS enforces no prerequisite | Not possible without AZ-104 | The exam fees are close. What differs is effort: Azure asks for a small amount of work every year, while AWS asks for a larger effort every three. ## What changed in AWS and Azure certifications in 2026? Microsoft retired associate exams and AWS is replacing its senior architect exam. Both vendors changed their catalogues this year, and study material written before the changes can mislead you. - **Microsoft retired several associate exams.** Its [retirement list](https://learn.microsoft.com/en-us/credentials/support/retired-certification-exams) shows AI-102 (Azure AI Engineer) retired on 30 June 2026 and AZ-204 (Azure Developer) on 31 July 2026. - **AWS is refreshing its senior architect exam.** Registration for Solutions Architect – Professional SAP-C03 opens on 27 October 2026, and the last day to sit SAP-C02 is 17 November 2026. - **AWS is also updating its machine learning associate exam**, with the last English sitting of MLA-C01 on 28 September 2026. ## How do you choose between AWS and Azure? Let local job postings decide. Open twenty job postings for the role you want, in the city you want, and count the cloud mentions. That is your answer. It beats every global statistic, including the ones in the table above, because you are not applying to the global market. You are applying to maybe forty companies. Regional and sector patterns are strong and stable. In India, large services companies, banks and insurers skew heavily Microsoft, often because the organisation already runs Active Directory and Microsoft 365, and the cloud decision followed. Product companies and funded startups skew AWS. Media, analytics and ML-heavy firms are where Google Cloud is genuinely competitive. | If most postings you see mention | Start with | Then | |---|---|---| | Azure, in a bank, insurer or services firm | AZ-104 | AZ-305 once you design solutions | | AWS, in a product company or startup | Solutions Architect – Associate | A deployed project, then the professional exam | | Both, roughly equally | The cloud your current employer uses | The second cloud a year later | | Neither, because the roles are Kubernetes-heavy | Certified Kubernetes Administrator | The cloud your employer adopts | ## How should you prepare for an AWS or Azure exam? Start from the official exam guide and finish with hands-on work in a real account; the sequence is the same for both vendors: 1. **Read the official exam guide.** Both vendors publish the domains and weightings, and they change them when exams update. 2. **Use the free official training.** Microsoft Learn training is free, and AWS Skill Builder offers more than 1,000 free learning resources. 3. **Build in a real account.** Deploy a small network, a compute tier and a database, then tear it down so the bill stays small. 4. **Take the official practice assessment.** Microsoft offers free practice assessments for many exams, and AWS offers official practice material through Skill Builder. 5. **Book the exam with a date four to eight weeks away.** A deadline turns vague study into a plan. ## What does neither certification prove? Neither certification proves you can fix what breaks at 2am. Both exams test whether you know which service solves which problem. Neither tests whether you can debug a networking issue between two subnets while a payment flow is down, which is what the job is. This is why the strongest candidates pair a certification with a small amount of visible operational evidence: a deployed architecture with infrastructure as code, a monitoring dashboard, and a written post-mortem of something you broke and fixed in your own account. That combination, one certification plus one real deployment, outperforms three certifications in every hiring conversation we have had on this subject. ## Common mistakes when choosing between AWS and Azure - **Studying both clouds at once.** Splitting attention early slows both down. - **Forgetting Azure's annual renewal.** The assessment is free, but a lapsed certification still looks careless. - **Assuming AZ-305 stands alone.** You need AZ-104 as well to earn the Expert certification. - **Studying from retired exam material.** Check the exam code on the vendor page before you buy a course. - **Skipping hands-on work.** Scenario questions on both tracks punish memorised answers. ## Where to go next Run the twenty-posting count this week, then book the exam for the cloud that wins with a date four to eight weeks out. The full ranking across all providers is in [Top 11 Cloud Certifications in 2026](https://www.1stepgrow.com/articles/top-cloud-certifications-2026). For AI credentials on either cloud, see our ranking of [generative AI certifications](https://www.1stepgrow.com/articles/best-genai-certifications-2026), and browse more head-to-heads on the [comparisons hub](https://www.1stepgrow.com/comparisons). ## Frequently asked questions ### Can I get both AWS and Azure certifications? You can, and platform engineers often do eventually. But sequentially, not in parallel — the second one is much easier once the underlying concepts are in place, and splitting attention across two clouds early slows both down. A sensible gap is a year of hands-on work on the first cloud before you start the second. ### Which is easier to pass? Azure's associate exams are generally considered slightly more approachable, leaning toward service knowledge. AWS leans harder on scenario reasoning. Neither is easy without hands-on time. Both vendors publish official practice material, so take a practice assessment early to find out which style suits you before you commit to a track. ### Does the certification expire? AWS certifications are valid for three years. Microsoft role-based certifications now require a free annual online renewal assessment, which is less demanding but more frequent. To renew an AWS certification, you pass the latest version of the exam or a higher-level one before it expires. Microsoft fundamentals certifications do not expire. ### Is Google Cloud certification worth considering instead? In data-heavy and ML-adjacent organisations, absolutely — GCP's Professional Data Engineer and Cloud Architect exams are excellent. In the general job market, the listing volume is lower, and Google Cloud professional certifications last two years rather than three. See our [cloud certification rankings](/articles/top-cloud-certifications-2026) for the full picture. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/aws-vs-azure-certification-path/). Cite with the title, "1stepGrow" and a link._ --- # Lists vs Tuples in Python: Differences and When to Use Each > Lists vs tuples in Python beyond 'one is mutable': the real differences in hashability, memory, safety and intent, plus a decision rule that works in practice. - **Author:** Ayushi Kulshreshta — AI Engineer (https://www.1stepgrow.com/authors/ayushi-kulshreshta/) - **Published:** Jul 25, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Comparison · **Read time:** 8 min - **Canonical URL:** https://www.1stepgrow.com/articles/lists-vs-tuples-python/ ## Key takeaways - Mutability is the root difference; hashability, memory and intent all follow from it. - The practical rule: homogeneous and growing means list; fixed-size record means tuple. - Tuples can be dictionary keys and set members. Lists cannot. - Performance differences are small enough that they should not drive the decision. The short answer on lists vs tuples in Python: lists are mutable and tuples are not. Use a list for a collection of similar items that will grow or change, and a tuple for a fixed record, a dictionary key or a value that must not change. Every other difference follows from mutability. "Tuples are immutable" still leaves traps. A tuple that holds a list cannot be a dictionary key, and a list used as a default argument quietly carries data from one function call to the next. This comparison is for Python learners who know both types exist but are unsure which to reach for. It covers hashability, memory, safety and intent, with a one-question decision rule. For everything you can do with tuples themselves, see the [Python tuples guide](https://www.1stepgrow.com/articles/python-tuples). | Feature | List | Tuple | | --- | --- | --- | | Mutable | Yes | No | | Hashable (usable as dict key / set member) | No | Yes, if contents are hashable | | Memory per element | Higher (over-allocates for growth) | Lower (exact size) | | Creation speed | Slower | Faster | | Iteration speed | Comparable | Comparable, marginally faster | | Can grow or shrink | Yes - append, extend, pop | No - concatenation makes a new tuple | | Sortable in place | Yes - .sort() | No - sorted() returns a list | | Typical use | A collection of similar things | One record with fixed fields | | Signals intent | This will change | This will not change | | Safe as a default argument | No - classic mutable default bug | Yes | **Verdict — List:** Lists are the right default for most code — the majority of sequences you build are homogeneous collections that grow, and mutability is what you want. Tuples earn their place in three specific situations: fixed-size records, keys for dictionaries and sets, and anywhere you want the language to enforce that something does not change. Choose on intent rather than on the small performance differences. ## What is the real difference between a list and a tuple? Mutability is the root difference, and three practical consequences follow from it. The Python design FAQ, [why are there separate tuple and list data types](https://docs.python.org/3/faq/design.html#why-are-there-separate-tuple-and-list-data-types), frames tuples like C structs and lists like arrays. **Hashability.** A hash must stay constant for the lifetime of the object, or dictionary lookups break. Since a list can change, it cannot be hashed. Since a tuple cannot change, it can — provided everything inside it is [hashable](https://docs.python.org/3/glossary.html#term-hashable) too. ```python valid = {(1, 2): "point"} invalid = {[1, 2]: "point"} # TypeError (unhashable type: 'list') # a tuple containing a list is also unhashable {(1, [2]): "x"} # TypeError (unhashable type: 'list') ``` **Memory.** A list over-allocates so that repeated `append` is amortised O(1). By contrast, a tuple allocates exactly what it needs. You can see this with [`sys.getsizeof`](https://docs.python.org/3/library/sys.html#sys.getsizeof): ```python import sys sys.getsizeof([1, 2, 3]) # 88 on 64-bit Python 3.14 sys.getsizeof((1, 2, 3)) # 72 on 64-bit Python 3.14 ``` These figures were measured as of September 2026 on Python 3.14. The exact numbers change between Python versions and platforms, but the tuple is consistently smaller. That is meaningful at scale — a million small tuples versus a million small lists is a real difference — and irrelevant otherwise. **Safety as a default argument.** This is the practical one, and the [Python programming FAQ](https://docs.python.org/3/faq/programming.html#why-are-default-values-shared-between-objects) covers it too: ```python # broken - the default list is created once and shared def add_item(item, basket=[]): basket.append(item) return basket add_item("a") # ['a'] add_item("b") # ['a', 'b'] <- surprise # correct def add_item(item, basket=None): if basket is None: basket = [] basket.append(item) return basket ``` Tuples, however, do not have this failure mode, because you cannot mutate them. ## How to decide between lists vs tuples in Python To decide, ask one question: **is this a collection of similar things, or one record with fixed fields?** A collection — user IDs, temperatures, rows to process — is a list. It is homogeneous, its length depends on the data, and you will probably append to it or sort it. On the other hand, a record — a coordinate, an RGB colour, a (name, age, city) triple — is a tuple. Its length is fixed by meaning, its positions have distinct roles, and changing it does not conceptually make sense. ```python # collection of similar things temperatures = [18, 25, 31, 12, 28] temperatures.append(30) # one record, fixed fields location = (19.0760, 72.8777) # a list of records cities = [ ("Mumbai", 19.0760, 72.8777), ("Delhi", 28.6139, 77.2090), ] ``` That last pattern — a list of tuples — is extremely common, and it shows the rule working in both directions at once. ## When to use a tuple vs a list: quick reference | Situation | Choose | Why | |---|---|---| | Rows you will filter, sort or append to | list | needs mutation | | A coordinate, colour or date parts | tuple | fixed fields | | A dictionary key or set member | tuple | must be hashable | | Several return values from a function | tuple | Python packs them automatically | | Constants such as allowed extensions | tuple | cannot be changed by accident | | A default argument value | tuple or `None` | avoids the shared-default bug | ## Where do tuples beat lists in real code? Tuples win in four situations: compound dictionary keys, set members, multiple return values and constants. Each is shown below with real output. **Compound dictionary keys:** ```python from collections import defaultdict transactions = [("North", "Q1", 1200.0), ("North", "Q1", 300.5), ("South", "Q1", 800.0)] sales = defaultdict(float) for region, quarter, amount in transactions: sales[(region, quarter)] += amount dict(sales) # {('North', 'Q1'): 1500.5, ('South', 'Q1'): 800.0} ``` **Set membership for pairs:** ```python edges = [(1, 2), (2, 1), (2, 3)] seen_edges = set() for a, b in edges: key = (min(a, b), max(a, b)) if key in seen_edges: continue seen_edges.add(key) sorted(seen_edges) # [(1, 2), (2, 3)] - the duplicate edge is skipped ``` **Returning several values:** ```python def describe(nums): return min(nums), max(nums), sum(nums) / len(nums) lo, hi, avg = describe([3, 1, 4, 1, 5]) # 1, 5, 2.8 ``` **Fixed configuration you do not want mutated:** ```python ALLOWED_EXTENSIONS = (".csv", ".json", ".parquet") "report.json".endswith(ALLOWED_EXTENSIONS) # True - endswith accepts a tuple ``` ## When is a list the right choice? Lists are right for anything you build up, filter, sort or modify: ```python scores = [("Ananya", 91), ("Rohit", 78), ("Meera", 85)] results = [] for name, score in scores: if score >= 80: results.append((name, score)) results.sort(key=lambda r: r[1], reverse=True) results # [('Ananya', 91), ('Meera', 85)] ``` Note that `sorted()` on a tuple returns a list, because it has to — there is no in-place sort for an immutable type. ## Common mistakes when choosing between lists and tuples - **Calling list methods on a tuple.** `(1, 2).append(3)` raises `AttributeError: 'tuple' object has no attribute 'append'`. - **Using a list as a dict key.** Convert it with `tuple()` first. - **Assuming a tuple is frozen all the way down.** A tuple holding a list can still change. - **Choosing tuples "for speed".** The gain is marginal; choose on intent. - **Using a list as a default argument.** Use `None`, or a tuple if the default is read-only. ## What should you use when tuple positions get hard to read? Use a `NamedTuple`. When `p[2]` no longer tells a reader what the value is, upgrade rather than tolerating it: ```python # unclear p = ("Ananya", 28, "Pune") print(p[2]) # clear from typing import NamedTuple class Person(NamedTuple): name: str age: int city: str p = Person("Ananya", 28, "Pune") print(p.city) # Pune ``` As a result, you keep tuple behaviour — immutability, hashability, unpacking — and gain names. Our [tuples guide](https://www.1stepgrow.com/articles/python-tuples) covers named tuples in more depth. ## Should you default to a list or a tuple? Use a list by default. Then reach for a tuple when you need a key, a record, or enforced immutability. In short, do not agonise over the performance difference in the lists vs tuples in Python debate. It is real, but it is almost never the thing that matters in your program. ## Related reading [Mastering Python tuples](https://www.1stepgrow.com/articles/python-tuples) goes deeper on tuple features. [Python basics](https://www.1stepgrow.com/articles/python-basics) covers the other core types, and more head-to-head breakdowns sit in the [comparisons hub](https://www.1stepgrow.com/comparisons). ## Frequently asked questions ### Are tuples faster than lists? Marginally, for creation, and they use less memory because they do not over-allocate. Iteration speed is essentially the same. The differences are small enough that they should almost never drive your choice, so pick on mutability and intent instead, and measure with timeit if a hot loop genuinely depends on it. ### Why can tuples be dictionary keys but not lists? Dictionary keys must be hashable, and hashability requires the value not to change. Otherwise the hash computed at insertion would no longer match after a mutation, and lookups would silently fail. Immutability is what makes tuples safe here, provided every element inside the tuple is hashable too, so a tuple holding a list still cannot be a key. ### Can I convert between lists and tuples? Yes, with list(my_tuple) and tuple(my_list). Both create a new object and copy the references to the same elements, so the conversion is O(n) and the elements themselves are not duplicated. A common pattern is to build a list while collecting data, then call tuple() once at the end to freeze the result. ### Why is a mutable default argument a bug? Default arguments are evaluated once, when the function is defined, not on each call. A default list is therefore shared across every call, and mutations accumulate from one call to the next. Use None as the default and create the list inside the function. A tuple default avoids the problem, because it cannot be mutated. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/lists-vs-tuples-python/). Cite with the title, "1stepGrow" and a link._ --- # Python Strings: F-Strings, Methods and the strip() Trap > Python strings in practice: f-strings, slicing, the methods worth memorising, why += loops slow down, and the encoding default that breaks scripts on Windows. - **Author:** Ayushi Kulshreshta — AI Engineer (https://www.1stepgrow.com/authors/ayushi-kulshreshta/) - **Published:** Jul 24, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/python-strings/ ## Key takeaways - F-strings are the default in modern Python — clearer and faster than format() or % formatting. - Strings are immutable, so building one in a loop with += can degrade to quadratic time. Use join(). - str.strip() removes any of the given characters, not a suffix — use removeprefix/removesuffix for that. - Always specify encoding='utf-8' when opening text files; the default depends on the platform before Python 3.15. Python strings are immutable sequences of Unicode characters: you can read, slice and search them, but every method hands back a new string. That one rule explains most string bugs, from `+=` loops that slow to a crawl to `"telecom.com".strip(".com")` returning `'tele'`. If you clean CSV columns, parse log lines or format reports, this guide covers the parts that cause real friction: f-strings and format specs, slicing, the methods worth memorising, and the encoding default that lets a script pass on Linux and fail on Windows. It follows [Python numeric data types](https://www.1stepgrow.com/articles/python-numeric-data-types) in the Python series. The full method list lives in the official [string methods reference](https://docs.python.org/3/builtins/stdtypes.html#string-methods). ## How do f-strings work? An f-string is a string literal prefixed with `f`, and anything inside braces is evaluated as an expression. It is the modern default, and the only formatting you need for new code: ```python name = "Ananya" score = 0.8734 count = 1_234_567 f"Hello, {name}" # 'Hello, Ananya' f"Score: {score:.1%}" # 'Score: 87.3%' f"Count: {count:,}" # 'Count: 1,234,567' f"Pi: {3.14159:.2f}" # 'Pi: 3.14' f"Hex: {255:#x}" # 'Hex: 0xff' f"Padded: {42:>8}" # 'Padded: 42' f"Centred: {'hi':^10}|" # 'Centred: hi |' ``` The codes after the colon come from the [format specification mini-language](https://docs.python.org/3/library/string.html#formatspec), which is worth bookmarking. Expressions also work inside the braces: ```python items = [1, 2, 3] f"There are {len(items)} items, total {sum(items)}" # 'There are 3 items, total 6' ``` Since Python 3.12, f-strings can reuse the enclosing quote character inside the braces, as described in [PEP 701's What's New entry](https://docs.python.org/3/whatsnew/3.12.html#pep-701-syntactic-formalization-of-f-strings). On older versions, this line is a syntax error: ```python cities = ["Mumbai", "Delhi"] f"Cities: {", ".join(cities)}" # 'Cities: Mumbai, Delhi' (3.12+) ``` The debugging form is genuinely useful and underused — a trailing `=` prints the expression alongside its value: ```python x = 42 f"{x=}" # 'x=42' f"{score * 100=:.1f}" # 'score * 100=87.3' ``` Multi-line output with alignment: ```python rows = [("Mumbai", 20.4), ("Delhi", 32.9), ("Pune", 7.4)] for city, pop in rows: print(f"{city:<10} {pop:>6.1f}M") ``` ```text Mumbai 20.4M Delhi 32.9M Pune 7.4M ``` Python 3.14 also added template strings (t-strings), which use a `t` prefix and return a `Template` object instead of a `str`. They exist for libraries that need to process the parts safely, such as HTML or SQL escaping; the [Python 3.14 release notes](https://docs.python.org/3/whatsnew/3.14.html#pep-750-template-string-literals) cover them. For everyday formatting, however, f-strings remain the tool. ## Which string formatting style should you use? | Style | Example | Use it for | |---|---|---| | f-string | `f"{name}: {score:.1f}"` | almost everything | | `str.format()` | `"{}: {:.1f}".format(name, score)` | templates stored apart from their data | | `%` formatting | `"%s: %.1f" % (name, score)` | reading legacy code and `logging` calls | ## Indexing and slicing Python strings Strings are sequences, so indexing and slicing work exactly as they do for lists: ```python s = "Python" s[0] # 'P' s[-1] # 'n' s[1:4] # 'yth' s[::-1] # 'nohtyP' - reversed s[10:] # '' - slices never raise IndexError ``` However, you cannot assign to an index, because strings are immutable: ```python s[0] = "J" # TypeError: 'str' object does not support item assignment "J" + s[1:] # 'Jython' - build a new string instead ``` ## Why is building a string with += slow? Because strings are immutable, every "modification" creates a new string and copies the old contents into it: ```python s = "hello" s.upper() # returns 'HELLO' print(s) # hello - the original is unchanged ``` Which makes this pattern potentially quadratic: ```python words = ["data", "science", "course"] # bad - O(n²) in general result = "" for word in words: result += word + " " # good - O(n) result = " ".join(words) # 'data science course' ``` With a thousand words the difference is invisible. With a million, it can be the difference between instant and a coffee break. CPython sometimes resizes the string in place for `+=`, so you may not see the slowdown locally, but that is an implementation detail you should not rely on. Therefore `join` remains the idiomatic answer whenever you are assembling a string from parts. For building with logic, collect then join: ```python records = [{"name": "a", "value": 1, "valid": True}, {"name": "b", "value": 2, "valid": False}] parts = [] for row in records: if row["valid"]: parts.append(f"{row['name']}: {row['value']}") output = "\n".join(parts) # 'a: 1' ``` ## Which string methods are worth memorising? A small set of methods covers most everyday text work: trimming, case changes, splitting and joining, searching, and testing what a string contains. ```python s = " Data Science Course " s.strip() # 'Data Science Course' s.lower() # ' data science course ' s.title() # ' Data Science Course ' s.replace("Course", "Program") # ' Data Science Program ' "a,b,c".split(",") # ['a', 'b', 'c'] "a b c".split() # ['a', 'b', 'c'] - splits on any whitespace run "a,b,c".split(",", 1) # ['a', 'b,c'] - limit the splits "hello".startswith("he") # True "file.csv".endswith(".csv") # True "abc" in "xxabcxx" # True "hello".find("l") # 2, or -1 if absent "hello".index("l") # 2, raises ValueError if absent "hello".count("l") # 2 ``` Case-checking and padding: ```python "abc123".isalnum() # True "123".isdigit() # True "abc".isalpha() # True " ".isspace() # True "5".zfill(3) # '005' "x".ljust(5, ".") # 'x....' ``` For quick reference, these are the methods you will reach for most often: | Method | What it does | Returns | |---|---|---| | `strip()` / `lstrip()` / `rstrip()` | removes a *set* of characters from the ends | new `str` | | `removeprefix()` / `removesuffix()` | removes one exact substring, once | new `str` | | `split()` / `join()` | breaks apart / glues together | `list` / `str` | | `replace(old, new)` | substitutes every occurrence | new `str` | | `find()` / `index()` | locates a substring | `int` (`-1` or `ValueError` if absent) | | `startswith()` / `endswith()` | checks the ends; also accepts a tuple | `bool` | | `casefold()` | aggressive lowercase for comparisons | new `str` | ## Why does strip() remove more than you expect? `strip` takes a **set of characters**, not a substring. The [`str.strip` documentation](https://docs.python.org/3/builtins/stdtypes.html#str.strip) says so directly, yet the name still misleads people: ```python "telecom.com".strip(".com") # 'tele' <- not what you wanted ``` It removed every leading and trailing character in `{'.', 'c', 'o', 'm'}`, so it kept eating past `.com` into `telecom`. With a name like `example.com` the bug stays hidden, because `e` is not in the set — which is exactly why it survives testing. Use the explicit methods instead: ```python "telecom.com".removesuffix(".com") # 'telecom' "www.example.com".removeprefix("www.") # 'example.com' ``` These were added in Python 3.9 and are the correct tool. Before that, the idiom was a conditional slice, which is why the bug was so common. ## How do you avoid encoding errors with Python strings? Encoding is the source of most real-world string pain in data work, and the fix is to be explicit. ```python # always be explicit with open("data.csv", encoding="utf-8") as f: text = f.read() # for messy sources that may contain invalid bytes with open("data.csv", encoding="utf-8", errors="replace") as f: text = f.read() ``` Without an explicit encoding, [`open()`](https://docs.python.org/3/builtins/functions.html#open) uses a platform-dependent locale encoding. For instance, code that works on your Linux CI can fail on a colleague's Windows laptop with a `UnicodeDecodeError` on a single accented character. [PEP 686](https://peps.python.org/pep-0686/) makes UTF-8 mode the default from Python 3.15. As of September 2026, however, the stable documentation still covers Python 3.14, so until every machine you support runs 3.15 or later, keep passing `encoding="utf-8"`. Encoding and decoding manually: ```python text = "café" data = text.encode("utf-8") # b'caf\xc3\xa9' - bytes data.decode("utf-8") # 'café' - back to str ``` The mental model: `str` is text, while `bytes` is what goes on disk or over the network. Encode to serialise, decode to read. Watch out for characters where visual length differs from `len`: ```python len("café") # 4 len("👨‍👩‍👧") # 5 - one visible glyph, several code points ``` If you are truncating user-facing text, this matters. ## How do you clean messy text columns? Chain a few methods in a small function: normalise odd whitespace, trim, then lowercase. This pattern shows up constantly in data work: ```python def clean(value: str) -> str: return ( value.replace(" ", " ") # non-breaking space to normal space .strip() .lower() ) raw_columns = [" Customer ID ", "Order Date"] columns = [clean(c) for c in raw_columns] # ['customer id', 'order date'] ``` Non-breaking spaces from spreadsheet exports are a genuine and infuriating cause of failed joins, because the values look identical and are not. Note the order, too: replace them before `strip()`, since the replacement can create new leading or trailing spaces. ## Common mistakes with Python strings - **Using `strip()` to remove a suffix.** Use `removesuffix()` instead. - **Forgetting that methods return new strings.** `s.upper()` on its own line does nothing useful; assign the result. - **Building long strings with `+=` in a loop.** Collect parts and `join` them. - **Opening files without `encoding=`.** The default differs between machines. - **Comparing user input case-sensitively.** Use `casefold()` on both sides. For anything more complex than fixed substitutions, use [regular expressions](https://www.1stepgrow.com/articles/regular-expressions-in-python). Next in the series, [Python operators](https://www.1stepgrow.com/articles/python-operators) covers `in`, `+` and `*` on strings alongside everything else. ## Related reading [Regular expressions Part 1](https://www.1stepgrow.com/articles/regular-expressions-in-python) covers pattern matching. [Python basics](https://www.1stepgrow.com/articles/python-basics) covers the core types. ## Frequently asked questions ### Why is += slow for building strings? Strings are immutable, so each += has to create a new string and copy both operands into it. Repeating that n times is O(n²) in the general case. CPython sometimes optimises the pattern in place, but you cannot rely on it across interpreters. Collect the pieces in a list and call ''.join(pieces) instead, which is O(n). ### What is the difference between strip and removeprefix? strip('abc') removes any leading or trailing characters that appear in the set {a, b, c}, repeatedly, until it meets a character outside the set. removeprefix('abc') removes exactly that one substring, once, if present. Using strip to remove a file extension or domain suffix is a classic bug, and removeprefix and removesuffix have existed since Python 3.9. ### Which string formatting should I use? Use f-strings for essentially everything in Python 3.6 and later. They are readable, fast, and support format specs and arbitrary expressions, and since Python 3.12 they can even reuse the same quote character inside the braces. Keep str.format() for templates defined separately from their data, and avoid % formatting in new code. ### How do I avoid encoding errors when reading files? Pass encoding='utf-8' explicitly to open(). Before Python 3.15 the default comes from the platform locale, so code that works on Linux can fail on Windows. For messy sources, errors='replace' substitutes a replacement character for unreadable bytes instead of raising UnicodeDecodeError, which lets you inspect the damage rather than crash. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/python-strings/). Cite with the title, "1stepGrow" and a link._ --- # Regular Expressions in Python, Part 1: Syntax, Functions, Flags > Regular expressions in Python from the ground up: raw strings, core syntax, match vs search vs fullmatch, flags, and five mistakes that silently break patterns. - **Author:** Ayushi Kulshreshta — AI Engineer (https://www.1stepgrow.com/authors/ayushi-kulshreshta/) - **Published:** Jul 23, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/regular-expressions-in-python/ ## Key takeaways - Always write patterns as raw strings, such as r'\d+', because Python processes backslashes in normal strings before the regex engine sees them. - re.match anchors at the start of the string and re.search looks anywhere, a difference that causes most beginner confusion. - Use re.fullmatch for validation, because search accepts a valid-looking fragment inside invalid input. - Quantifiers are greedy by default; add ? to make them lazy, or better, use a negated character class. - Pass flags by keyword, as flags=re.IGNORECASE, because re.sub and re.split treat a fourth positional argument as count. Regular expressions in Python live in the built-in `re` module, and they let you search, extract, validate and replace text by pattern rather than by exact string. Most real work needs only raw strings, a handful of syntax elements and four functions. Get one of those wrong, though, and a pattern fails without an error: `re.search("\bword\b", ...)` quietly returns `None`. This first part is for Python users who can write a loop but find regex write-only. It covers the syntax tables, which function to call when, the flags, a set of practical patterns, and the five mistakes behind most broken regex. [Part 2 covers advanced Python regex](https://www.1stepgrow.com/articles/advanced-python-regex): named groups, lookarounds and substitution with functions. ## Why should you always use raw strings? Start here, because it prevents a whole class of confusing failures. ```python import re # wrong - Python turns \b into a backspace character before regex sees it print(re.search("\bword\b", "a word here")) # None # right print(re.search(r"\bword\b", "a word here")) # ``` In a normal string, Python interprets backslash escapes first. `\d` happens to survive because it is not a recognised escape, which lulls people into thinking it is fine. Then `\b`, `\n` or `\t` silently changes meaning. Newer versions make the problem louder. According to the [What's New in Python 3.12](https://docs.python.org/3/whatsnew/3.12.html) notes, an invalid escape such as `"\d"` now emits a `SyntaxWarning`, and a future version will raise `SyntaxError`. As of September 2026, the current stable release, [Python 3.14.7](https://www.python.org/downloads/), still warns rather than fails. The [Regular Expression HOWTO](https://docs.python.org/3/howto/regex.html#the-backslash-plague) calls this "the backslash plague" and gives the same fix. **Write every pattern as `r"..."`.** It costs nothing. ## What is the core regex syntax? These tables cover most of what you will use day to day. The full list is in the [re module syntax reference](https://docs.python.org/3/library/re.html#regular-expression-syntax). **Character classes** | Pattern | Matches | |---|---| | `\d` / `\D` | A digit / any non-digit | | `\w` / `\W` | A word character (letters, digits, underscore) / anything else | | `\s` / `\S` | Whitespace / any non-whitespace | | `.` | Any character except a newline | | `[abc]` | Any one of a, b or c | | `[^abc]` | Any character except a, b or c | | `[a-z]` | Any character in the range | One detail most cheat sheets get wrong: for normal `str` patterns, `\w` and `\d` are Unicode-aware. `re.findall(r"\w+", "naïve café")` returns `['naïve', 'café']`, and `\d` matches Devanagari or Arabic digits too. Add `flags=re.ASCII` if you really mean only `[a-zA-Z0-9_]` and `[0-9]`. **Quantifiers** | Pattern | Meaning | |---|---| | `*` | Zero or more | | `+` | One or more | | `?` | Zero or one (optional) | | `{3}` | Exactly 3 | | `{2,5}` | Between 2 and 5 | | `{2,}` | 2 or more | **Anchors** | Pattern | Meaning | |---|---| | `^` | Start of string (or of each line, with `re.MULTILINE`) | | `$` | End of string, or just before a final newline (or end of each line, with `re.MULTILINE`) | | `\b` | Word boundary | ## How do match, search, findall and finditer differ? ```python text = "Order 12345 shipped on 2026-07-30, order 67890 pending" re.search(r"\d+", text) # re.match(r"Order", text) # re.match(r"\d+", text) # None - the string does not START with a digit re.findall(r"\d+", text) # ['12345', '2026', '07', '30', '67890'] re.finditer(r"\d+", text) # iterator of match objects, with positions ``` The `match` versus `search` distinction causes more confusion than anything else in the module. `match` is anchored at position zero, whereas `search` scans. The re documentation has a whole section on [search() vs. match()](https://docs.python.org/3/library/re.html#search-vs-match). **Use `search` unless you genuinely mean "must begin with".** `finditer` is the lazy option: it yields one match object at a time, so it suits long texts. The same idea is explained in [iterators and generators in Python](https://www.1stepgrow.com/articles/python-iterators-generators). Working with a match object: ```python m = re.search(r"(\d{4})-(\d{2})-(\d{2})", text) if m: print(m.group(0)) # 2026-07-30 - the whole match print(m.group(1)) # 2026 - first group print(m.groups()) # ('2026', '07', '30') print(m.span()) # (23, 33) - start and end positions ``` Always check for `None` before using the result. Calling `re.search(...).group()` on a non-match raises `AttributeError: 'NoneType' object has no attribute 'group'`, and it is the most common regex crash in production code. ## What is the difference between greedy and lazy quantifiers? A greedy quantifier takes as much text as it can, while a lazy one (written with a trailing `?`) takes as little as it can. Quantifiers are greedy by default: ```python html = "bold and italic" re.findall(r"<.+>", html) # ['bold and italic'] - one giant match re.findall(r"<.+?>", html) # ['', '', '', ''] - lazy ``` The `?` after `+` or `*` makes it lazy, so it matches as little as possible. This matters any time you are extracting between delimiters. Often, however, the better fix is to be more specific rather than lazy: ```python re.findall(r"<[^>]+>", html) # ['', '', '', ''] ``` `[^>]+` says "anything that is not a closing bracket", which cannot overshoot in the first place. Negated character classes are usually clearer than lazy quantifiers, and they give the engine less backtracking to do. ## Which regex flags should you know? Flags change how a pattern behaves. Pass them by keyword: | Flag | Effect | Example result | |---|---|---| | `re.IGNORECASE` | Case-insensitive matching | `findall(r"order \d+", text, flags=re.IGNORECASE)` finds both orders | | `re.MULTILINE` | `^` and `$` match at every line | `findall(r"^\w+", "one\ntwo", flags=re.MULTILINE)` gives `['one', 'two']` | | `re.DOTALL` | `.` also matches a newline | `findall(r"a.b", "a\nb", flags=re.DOTALL)` gives `['a\nb']` | | `re.ASCII` | `\w`, `\d`, `\s` match ASCII only | `findall(r"\w+", "café", flags=re.ASCII)` gives `['caf']` | | `re.VERBOSE` | Ignore whitespace, allow comments | See the readable pattern below | Combine flags with `|`, for example `flags=re.IGNORECASE | re.MULTILINE`. ## Practical examples of regular expressions in Python **Extracting numbers, including decimals and negatives:** ```python text = "Revenue was 45.2M, down -3.1% from 48.3M" re.findall(r"-?\d+(?:\.\d+)?", text) # ['45.2', '-3.1', '48.3'] ``` The `(?:\.\d+)?` part makes the decimal portion optional as a unit, so a trailing full stop in "version 3." is not swallowed. The `(?:...)` is a non-capturing group, which Part 2 explains. **Splitting on multiple delimiters:** ```python re.split(r"[,;|]\s*", "a, b; c|d") # ['a', 'b', 'c', 'd'] ``` **Validating a rough format:** ```python def looks_like_pincode(s: str) -> bool: return bool(re.fullmatch(r"[1-9]\d{5}", s)) looks_like_pincode("411001") # True looks_like_pincode("011001") # False - cannot start with 0 ``` `fullmatch` requires the entire string to match, which is what you almost always want for validation. Using `search` for validation is a classic mistake, because `re.search(r"\d{6}", "abc411001xyz")` succeeds. **Cleaning whitespace:** ```python re.sub(r"\s+", " ", "too much\n\nspace").strip() # 'too much space' ``` ## Should you compile regex patterns? Compile a pattern when you reuse it, mainly for readability rather than speed. If a pattern runs in a loop, compile it once: ```python pattern = re.compile(r"\b[A-Z]{2,}\b") pattern.findall("The API returned JSON via HTTP") # ['API', 'JSON', 'HTTP'] ``` The re documentation notes that Python caches the most recently used patterns, so the speed gain is modest. Even so, a compiled pattern gives you a named object, which reads better than a repeated literal. ## Readable patterns with re.VERBOSE Anything non-trivial should use `re.VERBOSE`, which ignores whitespace and allows comments: ```python date_re = re.compile(r""" (?P\d{4}) # four-digit year - (?P0[1-9]|1[0-2]) # 01 through 12 - (?P0[1-9]|[12]\d|3[01]) # 01 through 31 """, re.VERBOSE) m = date_re.search("due 2026-07-30") m.group("year") # '2026' m.groupdict() # {'year': '2026', 'month': '07', 'day': '30'} ``` Named groups make the extraction self-documenting. This version is longer than the cryptic equivalent and considerably easier to fix in six months. ## Common regex mistakes in Python - **Passing flags positionally.** `re.sub(r"cat", "dog", "Cat cat CAT", re.IGNORECASE)` returns `'Cat dog CAT'`, because the fourth positional argument is `count`, and `re.IGNORECASE` equals 2. Write `flags=re.IGNORECASE`, which returns `'dog dog dog'`. Since Python 3.13, the positional form also raises a `DeprecationWarning`. - **Validating with `$` instead of `fullmatch`.** `$` also matches just before a trailing newline, so `re.match(r"\d{6}$", "411001\n")` succeeds while `re.fullmatch(r"\d{6}", "411001\n")` does not. - **Assuming `\w` means ASCII.** It matches accented and non-Latin letters unless you add `re.ASCII`. - **Calling `.group()` without checking for `None`.** - **Forgetting the `r` prefix.** ## When should you not use regex? For genuinely structured formats, use a parser: `json`, `csv`, an HTML library or a date parser. Regex on HTML is the canonical example of a solution that works on your test cases and fails on real input. Email validation is another trap, because a fully correct pattern is notoriously enormous. Check for an `@` with something either side, then send a confirmation email. ## Related reading [Advanced Python regex (Part 2)](https://www.1stepgrow.com/articles/advanced-python-regex) covers groups, lookarounds and substitution. [Python strings](https://www.1stepgrow.com/articles/python-strings) covers the non-regex string methods, which are faster and clearer for fixed text. If a regex crash reaches production, [exception handling in Python](https://www.1stepgrow.com/articles/python-exception-handling) shows how to catch it narrowly. ## Frequently asked questions ### Why do regex patterns need the r prefix? Because Python processes backslash escapes in normal strings first. The sequence '\b' becomes a backspace character and silently breaks a word-boundary pattern. Unrecognised escapes such as '\d' still work today, but since Python 3.12 they trigger a SyntaxWarning and are due to become errors. A raw string passes backslashes through untouched, so always use r'...'. ### What is the difference between match, search and findall? re.match only checks at the start of the string, re.search scans for the first match anywhere, and re.findall returns every non-overlapping match as a list of strings. re.fullmatch requires the whole string to match. Use search unless you specifically mean 'must start with', and fullmatch when you are validating input. ### What does greedy mean in a regular expression? Quantifiers such as * and + take as much text as they can while still allowing the overall pattern to match. Adding ? after them makes them lazy, so they take as little as possible. This matters most when matching between delimiters like HTML tags or quotes, where a greedy .+ runs straight past the first closing delimiter. ### When should I not use regex? Avoid regex for structured formats that have real parsers, such as HTML, JSON, CSV and dates. Use json, csv, an HTML parser or datetime instead. Full email validation is another poor fit. Regular expressions are best for finding patterns in unstructured or semi-structured text, such as log lines, IDs and codes. --- _Source: 1stepGrow (https://www.1stepgrow.com/articles/regular-expressions-in-python/). Cite with the title, "1stepGrow" and a link._ --- # Advanced Python Regex: Groups, Lookarounds and Substitution (Part 2) > Advanced Python regex with tested output: named groups, lookarounds, re.sub with a function, and why a 24-character input made (a+)+$ run for 10 seconds. - **Author:** Ayushi Kulshreshta — AI Engineer (https://www.1stepgrow.com/authors/ayushi-kulshreshta/) - **Published:** Jul 22, 2026 · **Updated:** Sep 17, 2026 - **Topic:** Python & Programming · **Format:** Guide · **Read time:** 9 min - **Canonical URL:** https://www.1stepgrow.com/articles/advanced-python-regex/ ## Key takeaways - Named groups make patterns self-documenting and survive reordering, and groupdict() turns a match into a ready-made dictionary. - If a pattern contains capturing groups, re.findall returns the groups rather than the whole match. - Lookarounds assert without consuming, which is how you match context without capturing it; Python's lookbehind must be fixed width. - re.sub accepts a function as the replacement, which turns substitution into arbitrary transformation. - Nested quantifiers over overlapping patterns can backtrack catastrophically, so restructure the pattern or use an atomic group. Advanced Python regex means named groups that pull fields out of text, lookarounds that check context without capturing it, and `re.sub` with a function that transforms every match. Together they turn the `re` module into a small parser. They also bring a trap: the innocent-looking pattern `(a+)+$` took 10.7 seconds to reject a 24-character string in the timing measured below. This part is for readers who know the basics from [Part 1 on regular expressions in Python](https://www.1stepgrow.com/articles/regular-expressions-in-python): raw strings, flags, and `match` versus `search`. You will be able to parse log lines into dictionaries, match numbers only in the right context, rewrite dates and redact card numbers, and spot a pattern that can hang your program before it ships. ## How do capturing groups work? Parentheses capture the text they match, and you retrieve it by number with `group()` or all at once with `groups()`: ```python import re log = "2026-07-29 14:32:05 ERROR Database connection failed" m = re.search(r"(\d{4}-\d{2}-\d{2}) (\d{2}:\d{2}:\d{2}) (\w+) (.+)", log) m.group(1) # '2026-07-29' m.group(3) # 'ERROR' m.groups() # ('2026-07-29', '14:32:05', 'ERROR', 'Database connection failed') ``` However, positional groups break the moment you insert a group in the middle. Named groups do not: ```python pattern = re.compile( r"(?P\d{4}-\d{2}-\d{2}) " r"(?P