Follow one piece of data through any codebase.
Each lesson takes one real example and shows what goes in, what changes it and what comes out, one step at a time. The whole system stays on screen and lights up as you go. Zoom in, and click any part to see its files and data. No coding needed to follow along.
How does a language model turn text into its next character, and how does it learn to?
Watch vectors, attention and weights build up, then 3,000 training steps pull vowels together.
Numbers from our own tiny model that follows the repo's code.
Why does a chatbot answer the second question faster, when both start the same way?
A tree of saved prefixes splits, and memory rows are reused instead of recomputed: 2 tokens instead of 9.
Numbers from our own replay of the repo's cache rules.
What turns a language model into something that can fix a real bug?
A tower of context grows, turn by turn, through a refused edit to a one-line fix.
Real text from a recorded demonstration in the repo. No live model.
What does an AI tutor do with your question before it answers?
One textbook question through context, tools, retrieval and the loop that ends when the model stops asking.
Real prompt text. The conversation is an example.
What happens between pressing send and seeing the answer?
The hidden instructions, drawn to scale beside your short question.
Real prompt text. The conversation is an example.