AI Engineering from Scratch

Skip to content

AI / FROM SCRATCH

Contents
Catalog
Roadmap
Glossary

…

N

FIG_000 · curriculum v1.0 · 2026
open source · MIT

AI Engineeringfrom Scratch
435 lessons. 20 phases. Every algorithm built from raw math before a single framework gets imported.
Maintained by Rohit Ghumare and contributors. Run on your own machine.

How this works

Most AI material teaches in scattered pieces. A paper here, a fine-tuning post there, a flashy agent demo somewhere else. The pieces rarely line up. You ship a chatbot but can't explain its loss curve. You hook a function to an agent but can't say what attention does inside the model that's calling it.
This curriculum is the spine. 20 phases, 435 lessons, four languages: Python, TypeScript, Rust, Julia. Linear algebra at one end, autonomous swarms at the other. Every algorithm gets built from raw math first. Backprop. Tokenizer. Attention. Agent loop. By the time PyTorch shows up, you already know what it's doing under the hood.
Each lesson runs the same loop: read the problem, derive the math, write the code, run the test, keep the artifact. No five-minute videos, no copy-paste deploys, no hand-holding. Free, open source, and built to run on your own laptop.

Current Progress

Finished Lessons
bar
0 / 0

Phases
bar
0 / 0

Languages
bar
4

Glossary Terms
bar
···

Curriculum · 20 phases · 435 lessons
Tap a phase to expand its lessons. Each one ships when its math, code, and test are all written.

Complete
In progress
Planned

×

Progress saved in browser only
Reset progress

Colophon

The entire curriculum is on GitHub. Clone it, fork it, learn at your own pace. No paywall, no signup. Every lesson has runnable code in Python, TypeScript, Rust, or Julia, depending on what fits the concept best.

git clone https://github.com/rohitg00/ai-engineering-from-scratch.git
cp

© 2026 · open source · free forever

GitHub
Catalog
Glossary
Report

This curriculum, titled AI Engineering from Scratch, is designed as a comprehensive educational resource structured around twenty phases and four hundred thirty-five lessons, emphasizing the construction of every algorithm from fundamental mathematical principles rather than starting with pre-built frameworks. The central philosophy addresses the common issue in current AI education where material is often scattered, consisting of disconnected papers, fine-tuning posts, or superficial demonstrations, which fail to provide the necessary deep understanding of underlying mechanisms, such as the loss curves or internal model operations.

The curriculum functions as a rigorous spine, ensuring that concepts are built sequentially, linking essential concepts like backpropagation, tokenizers, attention mechanisms, and the agent loop directly to their mathematical basis. This grounding means that by the time practitioners encounter established frameworks like PyTorch, they possess a concrete understanding of the mechanics operating beneath the surface, rather than merely adopting existing tools blindly.

Each lesson adheres to a specific, reproducible instructional loop: the learner must first read the problem, derive the necessary mathematics, write the corresponding code, execute tests, and retain the resulting artifacts. This methodology intentionally excludes superficial elements such as short videos, copy-paste deployment scripts, or hand-holding, aiming for a method where mastery is achieved through direct derivation and implementation.

The curriculum is supported by a multi-language approach, utilizing Python, TypeScript, Rust, and Julia, allowing concepts to be implemented in the language best suited to the specific mathematical or computational requirement of the lesson. This approach ensures that the learning process is maximally flexible and applicable across different domains of AI engineering. The entire resource is free, open source, and designed to be run locally on a personal machine without requiring any sign-up or payment. The instructional content is maintained by Rohit Ghumare and various contributors, and the complete curriculum is available on GitHub for anyone to clone, fork, and study at their own pace.