Neural Networks: Zero to Hero
Builds neural networks and then a working language model from scratch in Python, one line at a time, with nothing hidden behind a framework. The clearest available answer to the question of what a model is actually doing.
Video
AI & Machine Learning
Andrej Karpathy
About 20 hours
Intermediate
Free
What it is
A free video series that starts with backpropagation on a single value and ends with a working transformer language model, writing every piece of it in front of you.
Why it works so well
Nothing is imported that has not been built. Frameworks are extraordinarily useful and they also hide exactly the parts you need to understand, so when a training run diverges you have no model of why. Building it up manually removes that opacity permanently.
What to expect
It is demanding and it is worth it. Type the code alongside rather than watching. The micrograd lecture on backpropagation and the lecture where the transformer is assembled are the two that change how you read every paper afterwards.
Best for: engineers who want to genuinely understand models rather than call them, and anyone considering machine learning work seriously.
Ready to start?
Opens on Andrej Karpathy in a new tab.
Stuck on something specific?
Writing only gets you so far. If you want an answer to your situation rather than the general case, book a session and we will work through it together. Every session is free; a few slots open each week.
Follow along
New writing, resources and project ideas land here first.