All roadmaps

AI and Data Scientist

From data to models you can defend.

The complete data-science path: math that matters, the Python stack, data wrangling and EDA, classical machine learning done honestly, deep learning, transformers and LLMs, and the portfolio that proves it.

Your progress0/57 topics · 0%

Progress is saved in this browser.

9 stages57 topics₹0 to follow

Filled nodes are the focus for this track; outlined ones can wait. Tap a topic to open it and mark it done, learning or skipped.

AI and Data Scientist in 2026Related roadmapsPython roadmapSQL roadmapAI Engineer roadmapData Analyst roadmapProve it as you goReading is half of it. This roadmap isbacked by a free, scored AI and DataScientist certification with a publicverification link and a ML Engineer jobsimulation on real tickets - evidence arecruiter can check, not a claim. Marktopics done, learning or skipped as yougo; your progress carries into theclassic view.Get certified - freeJob simulationClassic viewKnow the job before the toolsThe AI and Data Scientist job in2026Which skills to learnPython, SQL,statisticsML to LLMs,measuredAI-assisted analysisWhich role to prepare forDS vs ML eng vsAI engFresher vsswitcherSectors, cities and payWhat companies expectThe hiring barWhat interviewstestProof they verifyWhat gets you hiredA 90-day planProjects thatget readCertification and proofJust enough math, deeplyunderstood.Math That MattersLinear algebra essentialsCalculus intuitionProbabilityMake Python your native toolfor data.Python & The PyData StackPython fundamentalsNumPypandasJupyter & reproducibilityInterrogate a dataset beforemodelling it.Data Wrangling & EDAData cleaningExploratory data analysisVisualisationSQL for data scienceReason under uncertainty like ascientist.Statistics for InferenceSampling & estimationHypothesis testingExperiment designModel tabular data - still mostindustry ML.Classical Machine LearningRegressionClassificationGradient boostingUnsupervised learningValidation & leakageMetrics & imbalanceFeature engineeringNeural networks from intuitionto practice.Deep LearningNeural network fundamentalsPyTorchCNNs & RNNsTraining in practiceThe modern layer on top of thefundamentals.Transformers & LLMsAttention & transformersHugging Face ecosystemApplied LLMsMLOps basicsEvidence that survivesrecruiter scrutiny.Portfolio & ProofCase-study portfolioCompetitions & communityProve itFresher or experienced?Both tracks hold every topic. On the Fresher track thefoundations are filled and the later stages are outlined - dothem when a project needs them. On the Experienced track thefoundations are outlined and the later stages are filled. The2026 section at the top applies to both.Prefer a list?The classic view shows the same 8 stages as a sequentialchecklist with time estimates. Your marks are shared betweenthe two views.Open the classic view

Everything on this roadmap

Every stage and topic on the map, in reading order. Click any topic to open it.

1.Know the job before the tools

2.Just enough math, deeply understood.

3.Make Python your native tool for data.

4.Interrogate a dataset before modelling it.

5.Reason under uncertainty like a scientist.

6.Model tabular data - still most industry ML.

7.Neural networks from intuition to practice.

8.The modern layer on top of the fundamentals.

9.Evidence that survives recruiter scrutiny.

The AI and Data Scientist roadmap, topic by topic

Every topic on the map in reading order, with what to learn and the resources we recommend. Main topics are open; open any subtopic to read it. Select a topic name to open it in the topic panel, where you can track your progress.

  1. 1. Know the job before the tools

    The AI and Data Scientist job in 2026

    2026

    Data science in 2026 is three jobs that share a foundation: the data scientist who turns data into decisions and models, the machine learning engineer who ships models into products, and the AI engineer who builds on large language models. The US Bureau of Labor Statistics projects data scientist employment growing about 34 percent from 2024 to 2034, many times the average; the World Economic Forum's Future of Jobs 2025 report names big data specialists and AI and machine learning specialists among the fastest-growing roles; Naukri's JobSpeak reported AI and ML hiring in India up 49 percent year on year in February 2026. The four nodes answer which skills, which role, what companies expect and what gets you hired.

    Which skills to learn

    Python, SQL and statistics you can defend are the floor; machine learning learned with evaluation at the centre, then language models used and measured, is the 2026 layer; and AI-assisted analysis - using models to write, explore and summarise while verifying every claim - is the daily practice. The Stack Overflow 2025 survey shows Python rising seven points in a year, and Octoverse 2025 keeps it the second most used language on GitHub. The sub-nodes order the three.

    Python, SQL, statistics

    Python with pandas and the scientific stack, SQL with joins, window functions and aggregation, and statistics you can defend in a meeting - distributions, sampling, confidence, hypothesis tests, experiments and the mistakes that make a result wrong. The Stack Overflow 2025 survey recorded Python's largest jump in years, up seven points, and keeps SQL at 59 percent of developers; Octoverse 2025 has Python second on GitHub. The World Economic Forum's 2025 report finds analytical thinking the skill employers rank first. Learn these as tools for answering questions, not as syntax.

    ML to LLMs, measured

    Machine learning in 2026 is learned with evaluation at the centre: a baseline, a held-out set, the right metric for the business cost, leakage caught, and a model that beats the baseline by enough to matter. Then language models as a tool you call and measure - prompting, retrieval, structured outputs, an evaluation set - rather than something you train. The World Economic Forum names AI and big data the fastest-growing skills to 2030, and the Nasscom-Zinnov 2025 report found AI startups taking 91 percent of India's deep-tech funding. The AI Engineer roadmap carries the model-backed half in depth.

    AI-assisted analysis

    AI

    A data scientist now writes much of the analysis with an assistant: the pandas code, the SQL, the chart, the first draft of the memo. The discipline is verifying every number - a model will happily produce a plausible wrong join or an invented statistic. The Microsoft and LinkedIn Work Trend Index found 75 percent of knowledge workers using AI at work and 66 percent of leaders unwilling to hire without AI skills; Indeed's Hiring Lab reported in January 2026 that about 45 percent of data postings mention AI. The hire is the analyst who gets the speed and keeps the rigour: check the grain, the join, the maths, every time.

    Which role to prepare for

    Data scientist, machine learning engineer and AI engineer diverge on what you ship: insight, a trained model, or a model-backed feature. The sub-nodes compare them, lay out the fresher and switcher routes, and say where demand sits by sector and city.

    DS vs ML eng vs AI eng

    A data scientist answers questions and builds models that inform decisions; a machine learning engineer productionises models with pipelines, serving and monitoring; an AI engineer builds features on top of existing language models. The US Bureau of Labor Statistics projects data scientist jobs growing about 34 percent over the decade to 2034; LinkedIn's Jobs on the Rise 2026 ranked AI engineer the fastest-growing title in the United States with postings up 143 percent. The foundations are shared; choose by whether you want to find the answer, ship the model or build the product. The analyst path is the gentler entry and the Data Analyst roadmap covers it.

    Fresher vs switcher

    Freshers enter through analyst and junior data scientist roles, on the strength of statistics, SQL and one real project; HackerRank's 2025 report found junior hiring the hardest hit with 74 percent of developers struggling to land a job, so the project is the lever. Switchers from engineering, finance, operations or research bring domain data and should build on it - a model that improves a decision they used to make is the best portfolio piece there is. On this map freshers get the foundations filled and the advanced modelling outlined; experienced learners get the reverse.

    Sectors, cities and pay

    Demand is strongest in fintech, e-commerce, healthcare, SaaS and the global capability centres doing analytics and AI work from India; Naukri's JobSpeak showed AI and ML roles up 49 percent year on year in February 2026 while overall white-collar hiring grew 12 percent, and the Nasscom-Zinnov report counts more than 4,200 deep-tech startups in India with AI dominant. Bengaluru leads, then Hyderabad, Pune, Mumbai and Gurugram; remote roles for foreign companies pay more and ask for communication and evaluation discipline. Pay rises with models that moved a business number, not with the number of algorithms you can name.

    What companies expect

    Companies expect a data scientist to answer a business question with data honestly, build a model that is measured against a baseline, communicate the result to someone who will act on it, and use AI tools without letting them invent the numbers. HackerRank's 2025 report says practical assessments are the standard. The sub-nodes cover the bar, the interview and what gets verified.

    The hiring bar

    The bar is an answer a business can act on: the question stated precisely, the data understood at its grain, the analysis reproducible, the model measured against a baseline with the right metric, the limits stated, and the result communicated in a page and a chart. The Work Trend Index found 71 percent of leaders would prefer a less experienced candidate with AI skills over a more experienced one without; data teams expect you to use the tools for speed and to be the person who catches what they invented.

    What interviews test

    Expect a case study on real or realistic data - clean it, answer a question, build and evaluate a model, present it - often with an AI tool allowed; SQL questions with window functions; statistics questions about experiments, leakage and confidence; and a walk through your project. HackerRank's 2025 report says 66 percent of developers prefer practical assessments and that they are now the standard. The interviewer is listening for how you would know if you were wrong.

    Proof they verify

    Reviewers open the notebook or repository and read the evaluation: is there a baseline, a held-out set, a metric that matches the cost of a wrong answer, and a sentence on what the model cannot do. A project with that beats five tutorial datasets; a verified certification with a public link beats a course badge. OneRoadmap's AI and Data Scientist certification is free, scored and verifiable, and this roadmap's career stage names the external credentials employers recognise.

    What gets you hired

    A project that answers a real question with real data, a model with a documented evaluation, a readable repository, a verified certification and a clear account of how AI tools were used and checked. The sub-nodes give the ninety-day plan, the projects that get read and the proof that counts.

    A 90-day plan

    Month one: Python, SQL and statistics on one public dataset, ending in a short written analysis with a chart. Month two: a machine learning project with a baseline, a held-out evaluation and a model card, plus a language-model feature measured against an evaluation set. Month three: a deployed dashboard or notebook with the two projects, a readme on the decisions and the AI-assisted parts you verified, the AI and Data Scientist certification, and twenty targeted applications a week.

    Projects that get read

    Projects that get read start from a question: a churn or demand model with a baseline and a business-metric evaluation; an experiment analysis with power and confidence done properly; a retrieval-augmented question-answering tool over a real document set with an evaluation set and a cost note; a data-quality investigation that quantifies how much a problem moves a headline number. Each has a repository, a readme and a one-page memo. The World Economic Forum names big data specialists among the fastest-growing roles; a project that shows judgement with data is what that demand is for.

    Certification and proof

    Proof is verifiable: OneRoadmap's free AI and Data Scientist certification with a public link, a GitHub profile with real notebooks and code, a deployed project and a short account of how you used AI tools and what you checked. With over 180 million developers on GitHub in Octoverse 2025, the candidates who get replies are the ones whose evaluation a reviewer can read in a minute.

  2. 2. Just enough math, deeply understood.

    Math That Matters

    Just enough math, deeply understood.

    Time: about 3 weeks. Topics: Linear algebra essentials · Calculus intuition · Probability.

    Linear algebra essentials

    Optional on the Experienced track

    Vectors, matrices and dot products - the notation every model description is written in.

    Calculus intuition

    Optional on the Experienced track

    Derivatives and gradients as 'which way is downhill' - the idea behind all training.

    Probability

    Optional on the Experienced track

    Random variables, distributions, conditional probability and Bayes - the language of uncertainty.

  3. 3. Make Python your native tool for data.

    Python & The PyData Stack

    Make Python your native tool for data.

    Time: about 3-4 weeks. Topics: Python fundamentals · NumPy · pandas · Jupyter & reproducibility.

    Python fundamentals

    Optional on the Experienced track

    Functions, comprehensions, error handling and environments - clean code before clever models.

    NumPy

    Optional on the Experienced track

    Arrays, broadcasting and vectorised thinking - the substrate every ML library builds on.

    pandas

    Optional on the Experienced track

    Loading, cleaning, joining and reshaping real datasets - where data scientists spend most hours.

    Jupyter & reproducibility

    Optional on the Experienced track

    Notebooks for exploration, scripts for pipelines, seeds and environments for results others can rerun.

  4. 4. Interrogate a dataset before modelling it.

    Data Wrangling & EDA

    Interrogate a dataset before modelling it.

    Time: about 2-3 weeks. Topics: Data cleaning · Exploratory data analysis · Visualisation · SQL for data science.

    Build: EDA case study. Pick a public dataset; deliver a notebook with cleaning, five insights and honest caveats.

    Data cleaning

    Missing values, duplicates, outliers and type fixes - with the judgement calls documented.

    Exploratory data analysis

    Distributions, relationships and anomalies via plots and groupbys - the stage that decides whether modelling is even warranted.

    Visualisation

    matplotlib and seaborn fluency - plots that expose bad data before it fools your model.

    SQL for data science

    Joins, aggregation and window functions - most industry data still lives behind a SQL interface.

  5. 5. Reason under uncertainty like a scientist.

    Statistics for Inference

    Reason under uncertainty like a scientist.

    Time: about 2-3 weeks. Topics: Sampling & estimation · Hypothesis testing · Experiment design.

    Sampling & estimation

    Sampling distributions, standard errors and confidence intervals - what your point estimate is hiding.

    Hypothesis testing

    Nulls, p-values, power and multiple-comparison traps; why p-hacking ruins careers.

    Experiment design

    Randomisation, metrics and guardrails - designing A/B tests whose answers you can trust.

  6. 6. Model tabular data - still most industry ML.

    Classical Machine Learning

    Model tabular data - still most industry ML.

    Time: about 4-5 weeks. Topics: Regression · Classification · Gradient boosting · Unsupervised learning · Validation & leakage · Metrics & imbalance · Feature engineering.

    Build: End-to-end prediction model. A churn or price model: EDA, features, model comparison, error analysis and a business-facing summary.

    Regression

    Linear and regularised regression: coefficients, residuals and interpretation you can defend.

    Classification

    Logistic regression, decision trees, random forests, KNN and Naive Bayes - the standard cast and when each shines.

    Gradient boosting

    XGBoost/LightGBM - the tabular-data workhorses that win competitions and quietly run industry.

    Unsupervised learning

    k-means, hierarchical clustering and PCA - structure-finding when there are no labels.

    Validation & leakage

    Train/test discipline, cross-validation and the leakage patterns behind most 'amazing' models.

    Metrics & imbalance

    Precision/recall, ROC-AUC, calibration and cost-sensitive choices - the metric the business cares about, not the default.

    Feature engineering

    Encoding, scaling, interactions and domain features - better features beat fancier models.

  7. 7. Neural networks from intuition to practice.

    Deep Learning

    Neural networks from intuition to practice.

    Time: about 4-5 weeks. Topics: Neural network fundamentals · PyTorch · CNNs & RNNs · Training in practice.

    Neural network fundamentals

    Layers, activations, loss and backprop intuition - training as gradient descent on purpose.

    PyTorch

    Tensors, autograd, DataLoaders and the training loop - the research-to-production standard.

    CNNs & RNNs

    Convolutions for images, recurrence for sequences - the classic architectures and the intuitions that survive them.

    Training in practice

    Overfitting, regularisation, learning-rate schedules and transfer learning - the knobs that make models actually converge.

  8. 8. The modern layer on top of the fundamentals.

    Transformers & LLMs

    The modern layer on top of the fundamentals.

    Time: about 3-4 weeks. Topics: Attention & transformers · Hugging Face ecosystem · Applied LLMs · MLOps basics.

    Attention & transformers

    Optional on the Fresher track

    Self-attention, positional encoding and pretraining - what changed in 2017 and why it took over every modality.

    Hugging Face ecosystem

    Optional on the Fresher track

    Loading pretrained models, pipelines and fine-tuning on your own data.

    Applied LLMs

    Optional on the Fresher track

    Embeddings, RAG and prompt-based classification - combining foundation models with your own data products.

    MLOps basics

    Optional on the Fresher track

    Experiment tracking, model registries and monitoring drift - models as living software, not one-off notebooks.

  9. 9. Evidence that survives recruiter scrutiny.

    Portfolio & Proof

    Evidence that survives recruiter scrutiny.

    Time: about 2 weeks. Topics: Case-study portfolio · Competitions & community · Prove it.

    Case-study portfolio

    Two deep projects with written conclusions beat ten Kaggle notebooks; publish code and findings.

    Competitions & community

    Kaggle competitions and study groups - feedback loops that compound faster than solo study.

    Prove it

    Take the OneRoadmap AI & Data Scientist certification for verified, shareable proof.

Prove it as you go

Reading is half of it. This roadmap is backed by a free, scored AI and Data Scientist certification with a public verification link and a ML Engineer job simulation on real tickets - evidence a recruiter can check, not a claim. Mark topics done, learning or skipped as you go; your progress carries into the classic view.

Fresher or experienced?

Both tracks hold every topic. On the Fresher track the foundations are filled and the later stages are outlined - do them when a project needs them. On the Experienced track the foundations are outlined and the later stages are filled. The 2026 section at the top applies to both.

Prefer a list?

The classic view shows the same 8 stages as a sequential checklist with time estimates. Your marks are shared between the two views.

Frequently asked questions

What does a data scientist actually do?

A data scientist turns data into decisions and models: framing a business question, finding and cleaning the data, analysing it with statistics you can defend, building a model measured against a baseline, and communicating the result to someone who will act on it. In 2026 much of the code is drafted by an assistant and the scientist's job is the judgement - the question, the grain, the evaluation, the limits.

Data scientist, machine learning engineer or AI engineer?

They share a foundation and diverge on what you ship. A data scientist produces insight and models that inform decisions; a machine learning engineer productionises models with pipelines, serving and monitoring; an AI engineer builds features on top of existing language models. LinkedIn's 2026 Jobs on the Rise ranked AI engineer the fastest-growing US title; the analyst path is the gentler entry to all three.

Do I need a maths or statistics degree?

No, but you need statistics you can defend: distributions, sampling, confidence, hypothesis tests, experiments and the mistakes that make a result wrong. Linear algebra and calculus matter if you move into model research or fine-tuning. Most hiring tests judgement with data - leakage, baselines, the right metric - not proofs.

Python or R?

Python. It is the language of the data stack, of machine learning and of AI engineering, it is what assistants write best, and GitHub's Octoverse 2025 keeps it the second most used language on the platform. R remains strong in statistics, research and some finance teams; learn it when a job asks.

Should I learn machine learning before or after analysis?

After. Analysis - SQL, pandas, statistics, a clear memo - is where most data jobs start and where most data science work still happens. Machine learning learned with evaluation at the centre comes next; language models as a tool you call and measure come after that. The AI Engineer roadmap carries the model-backed half in depth.

How long does it take to become job-ready?

A beginner usually needs many months of steady practice to reach a project with a real question, a reconciled analysis and a model with a documented evaluation; a developer or analyst switching in needs less. Nobody can promise a timeline. The evidence decides: a project a reviewer can read, a certification they can verify and a memo a manager would act on.

Will AI replace data scientists?

Assistants now write the pandas code, the SQL and the first draft of the memo, and they invent numbers just as fluently. The parts that remain human - the question, the grain, the evaluation, catching the plausible wrong join, saying what the model cannot do, and taking responsibility for a decision - are the parts this roadmap spends the most time on.

How do I prove data science skills without job experience?

One project that starts from a question and shows the checks: a reconciled analysis, a model with a baseline, a held-out evaluation and a model card, a one-page memo and a note on what the assistant drafted and what you verified. Add OneRoadmap's free AI and Data Scientist certification with its public verification link and the Data Analyst job simulation, which a recruiter can check.