From Zero to Agents: A Foundational AI/ML Course, Built From First Principles

Most machine learning books ask you to trust the black box. This one doesn't.

From Zero to Agents rebuilds artificial intelligence from first principles — starting at the question of what a word even is to a computer, and ending with a working, tool-using AI agent you build yourself. Every concept is explained three ways: theoretically, mathematically, and in code — first written from scratch in raw Python, then reimplemented in PyTorch, so you always know exactly what the library is doing for you.

This is not a tutorial that skips the hard parts. Every equation is derived. Every result is verified by running the code — including the places where the first attempt didn't work, and the book says so.

Inside, you'll build:

• Tokenization and embeddings, from raw text to trainable vectors
• The mathematical toolkit — vectors, matrices, gradients, probability — explained through working code, not abstract proofs
• A complete neural network, with backpropagation derived and verified by hand
• Recurrent networks, LSTMs, and a full transformer with real multi-head attention, trained from scratch
• LoRA fine-tuning, derived and measured, not just described
• A working AI agent — tool use, memory, multi-agent coordination — arriving at the book's own title

Written for practitioners who want to open a real AI/ML research paper and read the mathematics fluently, on sight — not skim past it to the prose.