A free course · no accounts · no ads · no paywall
Learn AI from zero to the frontier.
Start knowing no math and no code. Finish by training your own GPT — in your browser. Every equation built from scratch, every idea something you can drag, run and break.
- 91
- lessons
- ~90h
- of hands-on work · 9 projects
- 0
- prerequisites
How it works
Not a wall of text. A lab.
Math from the first step
Never seen an equation? Perfect. Every symbol is introduced with a picture you can move, in plain words — from the number line up to gradients and attention.
A real TypeScript workbench
A full code editor with autocomplete and type-checking runs right in the page. You write every model yourself — no Python, no installs, no AI doing it for you.
Tiny models, trained live
Spam filters, neural nets, a name generator, a small GPT: all trained on your own device while you watch the loss fall. Nothing is a video.
Yours, everywhere, privately
Progress saves as you go and syncs across devices with a single key — no email, no password, no tracking. Free forever.
The course map
Eleven modules, one long descent.
Each module leans only on the ones before it, and Modules 1–9 each end with a project. Skip ahead if you already know something — every lesson works on its own — but if you start at zero, you’ll never hit a wall.
Honest pacing: about 90 hours of real work for a complete beginner. At five hours a week that’s roughly 4 months — about the same as a university course, without the tuition.
Start Here
What learning machines are, and how this course works.
2 lessons · ~55 min
Code from Zero
Just enough TypeScript to build every model in this course yourself.
8 lessons · ~8 h 15 min
Numbers & Functions
Math from the very first step — no prior knowledge assumed.
9 lessons · ~7 h 40 min
Vectors & Matrices
The language every neural network is written in.
7 lessons · ~7 h 15 min
Calculus: the Math of Change
How to know which way is downhill — the engine of learning.
7 lessons · ~6 h 40 min
Probability
Thinking clearly about uncertainty — how models express belief.
9 lessons · ~8 h 10 min
Machine Learning
Models, losses and training loops — the real thing, from scratch.
11 lessons · ~11 h 25 min
Neural Networks
Stack simple pieces into something that can learn anything.
11 lessons · ~12 h
Language Models
Teaching machines to predict the next piece of text.
8 lessons · ~9 h 10 min
Transformers
The architecture behind every frontier model, built piece by piece.
8 lessons · ~8 h 50 min
The Frontier
From raw text-predictor to the systems at the edge of the field.
11 lessons · ~9 h 10 min