Open-source MIT-licensed curriculum of 503 lessons across 20 phases that builds backprop, tokenizers, attention, and agent loops from raw math before importing frameworks.

Each lesson requires deriving math, writing code, and passing tests in Python, TypeScript, Rust, or Julia. The GitHub repo (47.3k stars) ships a six-volume EPUB/PDF book and four certification prep tracks with 295 practice questions.

Topics

Agent Engineering Deep Learning Transformer Architecture Large Language Models AI Engineering

Visit Site All Bookmarks

Discover Similar Content

William's Bookmark Library

/**/

AI Engineering from Scratch

aiengineeringfromscratch.comSaved August 20, 20264 min

via AI / FROM SCRATCH

Summary

Open-source MIT-licensed curriculum of 503 lessons across 20 phases that builds backprop, tokenizers, attention, and agent loops from raw math before importing frameworks.

Topics

Agent Engineering Deep Learning Transformer Architecture Large Language Models AI Engineering

Visit Site All Bookmarks