Diwali Offer!Flat 20% OFF
Enroll Now
PrakalpanaLive online tech training
Data Science⏱️ 12 min readπŸ“… Nov 12

Data Scientist Roadmap 2026: A Practical 6-Month Plan for India

AI
Ananya Iyerβ€’Senior Data Scientist
πŸ“‘ Contents (29 sections)

πŸ“ŒWhy a Roadmap Matters in 2026

Data science is still one of the best paid tech careers in India, but the job has changed. In 2026, hiring managers at product companies, GCCs and analytics firms expect a data scientist to do more than train a model in a notebook. They want someone who can pull data with SQL, clean it with pandas, reason with statistics, build and evaluate a model, explain it to a business team and, increasingly, use large language models where they genuinely help.

Most beginners fail not because the topics are hard, but because they learn them in the wrong order. They start with deep learning before they can write a GROUP BY, or collect certificates without a single project they can explain. This roadmap fixes the order. Follow it stage by stage and you will have the skills and the portfolio that interviewers actually screen for.

If you are coming from analytics, read the data analyst roadmap first. Data analysis is the foundation data science is built on.

πŸ“ŒStage 1: Python for Data Work

Python is the default language of data science. You do not need to be a software engineer, but you do need to be fluent.

What to learn

  • Core syntax: variables, data types, loops, conditionals, functions
  • Lists, dictionaries, sets, tuples and comprehensions
  • File handling, reading CSV and JSON
  • Modules, virtual environments and pip
  • Writing small reusable functions instead of copying cells
  • Jupyter notebooks and VS Code
  • How you know you are ready

    You can write a script that reads a messy CSV, cleans a few columns, computes summary numbers and saves the result, without searching every line. If Python itself is new, our Python programming course covers this stage live.

    πŸ“ŒStage 2: SQL, the Skill Everyone Underestimates

    In real jobs, most of your data lives in a warehouse, not a CSV. Nearly every data science interview in India has a SQL round, and many candidates lose offers here.

    What to learn

  • SELECT, WHERE, ORDER BY, LIMIT
  • GROUP BY with HAVING and aggregate functions
  • All join types and when each one duplicates rows
  • Subqueries and common table expressions (CTEs)
  • Window functions: ROW_NUMBER, RANK, LAG, LEAD, running totals
  • Date handling and CASE expressions
  • Practise on a real dataset such as e-commerce orders. Answer business questions: monthly revenue, repeat customers, top products per category, week over week growth.

    πŸ“ŒStage 3: Statistics and Probability

    Statistics is what separates a data scientist from someone who runs model.fit(). You need intuition more than formulas.

    Core topics

  • Descriptive statistics: mean, median, variance, standard deviation, percentiles
  • Distributions: normal, binomial, Poisson, and why skew matters
  • Sampling, the central limit theorem and confidence intervals
  • Hypothesis testing: null hypothesis, p-values, type I and type II errors
  • t-tests, chi-square tests and when to use each
  • Correlation versus causation
  • A/B testing design: sample size, power, and avoiding peeking
  • Basic Bayesian thinking
  • Linear algebra and calculus help later for deep learning, but for most applied roles you only need vectors, matrices, dot products and the idea of a gradient.

    πŸ“ŒStage 4: pandas, NumPy and Data Cleaning

    Real data is messy. Expect to spend most of your project time here.

  • NumPy arrays and vectorised operations
  • pandas DataFrames: selecting, filtering, groupby, merge, pivot tables
  • Handling missing values, duplicates and outliers
  • Working with dates and categorical data
  • Reshaping data with melt and pivot
  • Writing clean, repeatable cleaning steps
  • πŸ“ŒStage 5: Visualisation and Storytelling

    A model nobody understands never ships. Learn to show what the data says.

  • Matplotlib and Seaborn for exploratory charts
  • Plotly for interactive charts
  • Choosing the right chart: distributions, comparisons, trends, relationships
  • Exploratory data analysis (EDA) as a habit before modelling
  • Presenting findings to non-technical stakeholders in three slides or fewer
  • πŸ“ŒStage 6: Machine Learning with scikit-learn

    This is the core of the job. Learn the concepts deeply, then the library.

    Supervised learning

  • Linear and logistic regression
  • Decision trees, random forests
  • Gradient boosting with XGBoost, LightGBM or CatBoost
  • k-nearest neighbours and support vector machines
  • Unsupervised learning

  • k-means and hierarchical clustering
  • PCA for dimensionality reduction
  • The workflow that matters more than any algorithm

  • Train, validation and test splits, and cross validation
  • Feature engineering and encoding
  • Pipelines so preprocessing never leaks into the test set
  • Metrics: accuracy, precision, recall, F1, ROC AUC, RMSE, MAE
  • Handling class imbalance
  • Hyperparameter tuning
  • Model interpretation with feature importance and SHAP
  • Gradient boosting on tabular data is what most Indian companies actually use in production for churn, fraud, credit risk and demand forecasting. Get very good at it.

    πŸ“ŒStage 7: Deep Learning Basics

    You do not need to become a research scientist, but you should understand neural networks.

  • Neurons, layers, activation functions, loss and backpropagation
  • PyTorch basics: tensors, a training loop, a simple network
  • CNNs for images and the idea of transfer learning
  • Sequence models and the transformer architecture at a high level
  • Using pretrained models from Hugging Face
  • πŸ“ŒStage 8: GenAI for Data Scientists

    In 2026, data scientists are expected to know where large language models fit and where they do not.

  • Calling LLM APIs and writing reliable prompts
  • Embeddings and semantic search
  • Retrieval augmented generation (RAG) basics
  • Using LLMs for text classification, extraction and summarisation
  • Evaluating LLM output instead of trusting it
  • Cost, latency and privacy trade-offs
  • The skill employers value is judgement: knowing when a simple logistic regression beats an expensive LLM call.

    πŸ“ŒStage 9: MLOps Basics

    A model that lives only in a notebook has no business value.

  • Git and clean project structure
  • Saving models and tracking experiments with MLflow
  • Serving a model behind a FastAPI endpoint
  • Docker basics to package your model, see our Docker guide
  • Monitoring for data drift and model decay
  • πŸ“ŒStage 10: Projects and Portfolio

    Two strong projects beat ten tutorial clones. Pick problems with business meaning.

    Project ideas that impress interviewers

  • Customer churn prediction for a telecom or SaaS dataset, with a cost based threshold and a short business recommendation
  • Demand forecasting for retail sales with proper time series validation
  • Credit risk scoring with explainability using SHAP
  • Support ticket classifier that compares a classic ML model with an LLM approach on accuracy and cost
  • For each project, publish a clean GitHub repo with a README covering the problem, data, approach, results and what you would do next. Deploy at least one as a small web app.

    πŸ“ŒThe 6-Month Plan

    MonthFocusOutput

    1Python and SQL50 SQL problems solved, a cleaning script 2Statistics, pandas, EDAAn EDA report on a real dataset 3Machine learning coreA churn model with a proper pipeline 4Advanced ML and boostingForecasting or credit risk project 5Deep learning and GenAIAn LLM or RAG mini project 6MLOps, portfolio, interviewsDeployed project, resume, mock interviews

    Working professionals can stretch this plan by a couple of months. Consistency matters more than speed: an hour a day beats a weekend binge.

    πŸ“ŒYour Data Science Toolkit for 2026

    AreaTools to know

    Language and notebooksPython, Jupyter, VS Code DataSQL, pandas, NumPy, Polars for larger data VisualisationMatplotlib, Seaborn, Plotly, Power BI or Tableau basics Machine learningscikit-learn, XGBoost, LightGBM, SHAP Deep learningPyTorch, Hugging Face Transformers GenAILLM APIs, embeddings, a vector database MLOpsGit, MLflow, FastAPI, Docker CloudOne of AWS SageMaker, GCP Vertex AI or Azure ML

    You do not need every tool on day one. Go deep on Python, SQL, pandas and scikit-learn first, then add the rest as your projects need them.

    πŸ“ŒCommon Mistakes to Avoid

  • Collecting certificates instead of building projects
  • Skipping SQL and statistics because they feel boring
  • Using Kaggle accuracy as the only measure of a project
  • Not being able to explain why you chose a metric
  • Ignoring data leakage, which silently inflates results
  • πŸ“ŒFAQ

    Can I become a data scientist without a maths degree?

    Yes. You need applied statistics and intuition, not a degree. Many working data scientists in India come from engineering, commerce and science backgrounds.

    Is data science still worth it with AI tools everywhere?

    Yes. AI tools speed up coding, but companies still need people who frame problems, pick the right data, validate results and explain decisions. That judgement is the job.

    Should I learn Python or R?

    Python. It dominates industry hiring in India and connects naturally to ML engineering and GenAI.

    Do I need a GPU?

    No. Free cloud notebooks are enough for learning deep learning basics.

    What should I learn next?

    Prepare with the data science interview questions and check the data scientist salary guide.

    πŸ“ŒLearn It Live

    Prakalpana's Data Science course follows this exact roadmap and is taught live online with an optional 1-on-1 track, real projects, code reviews and mock interviews. Explore the full data science track, or WhatsApp or call +91 9180377502 for a free demo.

    AI

    Written by

    Ananya Iyer

    Senior Data Scientist

    πŸš€ Master Data Science

    Join 5000+ developers

    Not sure what to do? Book career counselling (β‚Ή200 / 30 min)

    Not sure what to learn next?

    A 1-on-1 session with an industry mentor and a roadmap built for you. How it works

    1. 1Discuss your goals
    2. 2Get a personal plan
    3. 3Start learning
    Book for β‚Ή200/ 30 min