Learn AI from scratch, by doing
Most AI courses ask you to watch hours of video before you write a single line of code. Here it is the other way round: every concept starts with a concrete problem that you solve in your browser, and an automatic judge tells you right away whether your solution holds up. You start from Python basics and build up, step by step, to neural networks, computer vision and language models. It is free and nothing needs installing.
Why learning by solving problems works
You remember what you have had to figure out. Each exercise puts you in front of a situation where the concept is needed (cleaning messy data, predicting a price, recognising a digit) and lets you find the idea before naming it. The judge runs your code on hidden tests, so a solution that only looks right does not pass. Solutions unlock once you have solved the exercise, and every student pledges not to use generative AI on the exercises: your progress reflects what you actually know.
Six levels, from your first line of Python to transformers and beyond
The path is split into six levels. The first covers Python, NumPy and data manipulation; the second classical machine learning (regressions, trees, clustering, how to evaluate a model honestly); the next ones neural networks, computer vision and natural language processing, up to transformer-based models, and a sixth, frontier level opens onto ML systems, post-training and research papers. Each level ends with a boss challenge that checks you have really mastered it, and strong students can skip ahead by beating it. Chapter by chapter:
Level 1
Python: the basics · NumPy & vectors · Pandas & visualization · Probability · Beginner algorithms (optional)
Level 2
Statistics · Supervised models 1 · Supervised models 2 · Introduction to unsupervised learning · Evaluation methodology · Data preparation & feature engineering · Complete pipeline · How a machine learns
Level 3
Linear algebra for ML · Computing a derivative (bonus) · Optimization · Loss functions & information theory · Neural networks · Backpropagation · Training with PyTorch · Intermediate algorithms (optional)
Level 4
Vision · Text & embeddings · Sequences & encoder-decoder · Audio · Finetuning 1 · Few labels, shifting data · Mixtures, anomalies, matching
You do not learn alone
588 students from 89 countries train here and have solved 8,330 exercises so far. A daily question, a timed revision dojo, challenges between friends and a help forum where those who solved an exercise help those still stuck keep practice regular.
Going further: the AI olympiads
The platform was built by coaches of the International Olympiad in Artificial Intelligence (IOAI), and its upper levels are calibrated on olympiad difficulty. If you want to aim for a national team, the same path takes you there. How to prepare for the IOAI
Frequently asked questions
Can I learn AI with no prior knowledge?
Yes. The path starts with Python basics, then NumPy and data manipulation, before any machine learning. A 2-minute placement test suggests where to start if you already know some of it.
Do I need to install anything?
No. Exercises run in the browser: you write Python, an automatic judge runs it on hidden tests and gives you a verdict within seconds. The heaviest deep learning exercises use a free Colab notebook.
How much maths do I need?
High-school maths is enough to start. The notions you need later (vectors, matrices, derivatives, probabilities) are introduced through problems when the curriculum reaches them.
Is it really free?
Yes, training is free for students. Create an account with your email or Google and start.
How long does it take?
It depends on your pace. Each level takes a few weeks of regular practice; the daily question and the revision dojo help you keep going a few minutes a day.
Start now
Creating an account takes a minute. Take the placement test or start straight from the first chapter.