Short answer: To learn data science from scratch, start with Python and basic statistics, then learn to clean and explore data with NumPy and pandas, build machine learning models with scikit-learn, and finish with deep learning and NLP. Generative AI (large language models and RAG) and agentic AI (systems that use models with tools to complete tasks) come last, because they build on everything before them. Along the way, complete three to five projects that each answer a real question. No prior coding is required, but expect several months of regular practice.
This guide is for students, graduates, data analysts and working professionals who want a clear learning order, including where Gen AI and agentic AI fit. It covers what data scientists do, a step-by-step plan, projects to build and the roles you can aim for.
What is data science?
Data science is the practice of using statistics, programming and machine learning to turn raw data into insights, predictions and better decisions. Every project starts with a question, such as “which customers are likely to leave?”, and ends with an answer people can act on.
It helps to see how the newer topics fit in. Each layer builds on the one below it:
| Layer | What it does | Example |
|---|---|---|
| Data science | Analyses data to explain and support decisions | Which regions drove last quarter’s drop in sales? |
| Machine learning | Learns patterns from past data to predict or classify | Will this customer cancel next month? |
| Generative AI | Creates new content such as text, code or images | Draft a personalised retention email for this customer |
| Agentic AI | Uses models with tools, memory and steps to complete a task | Find at-risk customers, draft emails and queue them for human review |
Agentic AI is not simply “better AI”. Anthropic describes agents as systems where a language model directs its own process and tool use to accomplish a task, and recommends starting with the simplest solution that works (Anthropic: Building effective agents). Most real AI systems still depend on clean data, sound statistics and careful evaluation, which is why data science comes first.
Do you need maths or coding to start?
You do not need to be a programmer or a mathematician to begin. You need school-level maths, a willingness to practise, and patience with error messages. You will learn Python from the basics, and the statistics you need (averages, spread, probability, distributions and hypothesis tests) as you go. Linear algebra and calculus help you understand how models work, but you can learn the intuition first and the detail later.
How to learn data science, step by step
- Python basics. Variables, data types, loops, functions, lists and dictionaries, and reading files. Ready when you can write a short program that reads a CSV file and prints a summary.
- Statistics and probability. Descriptive statistics, distributions, sampling, correlation, and hypothesis testing. Ready when you can explain whether a difference between two groups is likely to be real or just noise.
- NumPy and pandas. Load, clean, filter, group and join data (NumPy for beginners, pandas: Getting started). Ready when you can take a messy dataset and produce a clean table ready for analysis.
- Exploratory analysis and visualisation. Charts that reveal patterns, outliers and relationships. Ready when you can present three findings from a dataset, each with a supporting chart.
- Machine learning. Regression, classification, clustering, train-test splits, cross-validation and evaluation metrics, using scikit-learn (scikit-learn: Getting started). Google’s Machine Learning Crash Course is a useful free companion. Ready when you can build a model, compare it with a simple baseline and explain its errors.
- Deep learning and NLP. Neural networks with TensorFlow and Keras, and text processing for tasks such as sentiment analysis. Ready when you can train a small text classifier and explain how you measured it.
- Generative AI. How large language models work at a high level, prompting, embeddings, and retrieval-augmented generation (RAG), where a model answers using documents you supply. Ready when you can build a chatbot that answers questions from a set of PDFs and cites them.
- Agentic AI. Combine models with tools, memory and multi-step workflows, with clear limits and human checks. Ready when you can build a small agent that uses two tools to finish a task, and explain how you stop it from doing the wrong thing.
- Pythonand statistics
- DataNumPy, pandas
- MLscikit-learn
- Deep learningKeras, NLP
- Gen AILLMs, RAG, agents
Projects that prove your skills
Projects matter more than certificates when you apply for jobs. Use public datasets, keep each project small and finished, and write up the question, method, result and limitations.
- Exploratory analysis: clean a public dataset and present three findings with charts.
- Prediction model: predict a yes-or-no outcome, such as customer churn, and compare your model with a simple baseline.
- Text project: classify reviews or support tickets by sentiment or topic.
- RAG chatbot: answer questions from a set of documents, showing the source for each answer.
- Small AI agent: automate a short multi-step task with tools, and document the safety checks you added.
Data science and AI job roles
| Role | Focus | Typical starting point |
|---|---|---|
| Data analyst | Reports, dashboards and explaining what happened | SQL, Excel, Power BI or Tableau, some Python |
| Data scientist | Statistical analysis and predictive models | Python, statistics, machine learning |
| Machine learning engineer | Putting models into production and keeping them running | Strong Python, ML, software engineering |
| AI / GenAI engineer | Building applications with LLMs, RAG and agents | Python, APIs, ML fundamentals, evaluation |
Many people start as data analysts and move into data science as their skills grow. If that route appeals to you, see how to become a data analyst.
Using AI responsibly
Models can be wrong, biased or confidently invent answers. Good practitioners test their models on data they have not seen, check results for different groups of people, protect personal data, and keep a human in the loop for important decisions. The NIST AI Risk Management Framework is a useful reference for thinking about these risks (NIST AI RMF).
Common mistakes beginners make
- Jumping straight to Gen AI. Without data handling and evaluation skills, it is hard to build anything reliable.
- Copying notebooks without understanding them. Rebuild each step yourself and explain why it is there.
- Chasing accuracy alone. Always compare against a simple baseline and check what the model gets wrong.
- Skipping statistics. It is what separates a real finding from a lucky pattern.
- Too many unfinished projects. Three finished, well-explained projects beat ten half-done ones.
Learning data science at ASTI
ASTI’s Data Science with Agentic AI & Gen AI course follows this order: Python, statistics, machine learning, deep learning and NLP, then generative AI with LLMs, RAG and AI agents. It includes 120+ hours of instructor-led training over about five to six months, case studies in every module, GenAI projects such as a RAG chatbot and AI agents, and a real-time project. You can learn in the classroom at BTM 1st Stage, Bangalore, or live online. The course includes placement and career support; placement assistance does not guarantee a job.
Still comparing options? See which software course to choose.
Frequently asked questions
How long does it take to learn data science?
It depends on your background and weekly practice time. As a reference, ASTI’s course runs for about five to six months, with 120+ hours of instructor-led training plus assignments and projects. Building a strong portfolio usually takes extra time beyond that.
Can I learn data science without a computer science degree?
Yes. People move into data science from engineering, commerce, science, economics and other fields. What employers look for is evidence that you can work with data: your projects and how well you explain them.
Should I learn Gen AI before machine learning?
No. You can experiment with Gen AI tools early, but to build reliable AI applications you need data handling, machine learning basics and evaluation skills first.
What is the difference between generative AI and agentic AI?
Generative AI creates content, such as text or code, in response to a prompt. Agentic AI uses those models together with tools, memory and multi-step plans to complete a task, ideally within limits set by people.
Is Python enough for data science?
Python is the main language for most data science and AI work. You will also need SQL to get data from databases, and statistics to interpret results correctly.
Sources
- Building effective agents, Anthropic
- NumPy: the absolute basics for beginners, NumPy documentation
- pandas: Getting started, pandas documentation
- scikit-learn: Getting started, scikit-learn documentation
- Machine Learning Crash Course, Google for Developers
- AI Risk Management Framework, NIST
- Course details: the Data Science with Agentic AI & Gen AI course page, checked 29 Oct 2026