The path
Ten tracks, one story
Start at tokens and probability, end with a production video-search system. Each track builds on the one before it.
Curriculum
Every lesson, in one tree
Tracks branch into topics, topics into lessons. Search to jump straight to a subject.
No lessons match that search.
About
Hi, I'm Mehdi.
I'm a research scientist at Meta, working on computer vision, computational photography and generative AI.
I learn best when an idea comes with a small piece of code I can run and change. This site is built the same way: each concept is explained with its math, a minimal implementation, and a figure you can interact with.
The LLM & VLM course is the main project here — a complete path from the first token to deploying multimodal models.