rohitg00/ai-engineering-from-scratch

Learn it. Build it. Ship it for others

Summary

ai-engineering-from-scratch is an MIT-licensed open curriculum of 523 lessons across 20 phases that teaches AI engineering end-to-end, spanning linear algebra, classical ML, deep learning, transformers, LLMs, agents, MCP, and production deployment. Lessons ship in Python, TypeScript, Rust, and Julia, each with runnable code, narrative docs, and a reusable artifact such as a prompt, skill, agent, or MCP server. The curriculum is positioned as a learning spine rather than a library, using a "Build It / Use It" pattern where engineers implement algorithms from raw math before invoking frameworks like PyTorch.

What it is useful for

This is useful when an engineer wants a structured, self-paced path that ties theory to applied AI instead of scattered blog posts or paper summaries. It fits four scenarios documented in the README: (1) beginners starting with Phase 0 (Setup and Tooling) and Phase 1 (Math Foundations); (2) Python-literate engineers moving into LLM and agent work via Phase 11 (LLM Engineering) and Phase 14 (Agent Engineering); (3) engineers adopting the Model Context Protocol, with a focused 17-lesson route covering stateless requests, transports, security, and conformance evidence; and (4) certification candidates, since the repo includes prep tracks for four Claude certification tracks and the 34-lesson MCP Associate (MCPA) curriculum. It is less useful if you only want a single library or framework integration — the value comes from following the ~342-hour path.

How engineers can use it

The README documents three entry points. The fastest is to install tutor skills into a coding agent:

npx skills add rohitg00/ai-engineering-from-scratch

Then invoke start-learning (Codex), /start-learning (Claude Code), or the skill by name in another compatible host. This creates LEARNING.md driven by a placement quiz.

For engineers who want to run lesson code directly, the documented path is:

git clone https://github.com/rohitg00/ai-engineering-from-scratch.git
cd ai-engineering-from-scratch
python3 phases/00-setup-and-tooling/01-dev-environment/code/verify.py --route beginner
python3 phases/01-math-foundations/01-linear-algebra-intuition/code/vectors.py

Every lesson folder follows the same structure: code/ for runnable implementations, docs/en.md for narrative, and outputs/ for the shipped artifact. The README instructs engineers to capture terminal output as "evidence" and explain it before moving on.

A third option is reading lessons on the website with no setup. Documented limitations: the certification tracks are independent study material — they are not affiliated with Anthropic or the Agentic AI Foundation and cannot guarantee passing scores — and full tutor functionality requires Node.js, npx, python3, a supported host, and a writable skill scope.