← ResourcesFree Courses

AI Engineering From Scratch: The Free 511-Lesson GitHub Curriculum (Beginner Roadmap)

A complete AI engineering curriculum — 511 lessons, 20 phases, ~330 hours — free on GitHub with no signup. Math through mini-GPTs, AI agents, and MCP servers, with nearly every lesson shipping a portfolio artifact. The link plus the exact beginner path by goal.

An AI engineering degree, sitting free on GitHub

AI Engineering from Scratch is a complete curriculum: 511 lessons across 20 phases — roughly 330 hours by the repo's own count — running from math and classical machine learning all the way to pre-training a mini-GPT, building AI agents, and shipping MCP servers. In Python, TypeScript, Rust, or Julia.

And it's actually free. MIT license, no paywall, no signup, no email wall — even the 6-volume book edition is a free download. It's sitting at 50,000+ GitHub stars, built by Rohit Ghumare, a Google Developer Expert whose previous project earned 27K stars.

💻 Get the free curriculum →

github.com/rohitg00/ai-engineering-from-scratch · MIT · no signup

The trap with a 330-hour curriculum: treating it like a Netflix series you'll "get through." Nearly every lesson ships a reusable artifact — a prompt, a skill, an agent, or an MCP server you keep — and the whole game is collecting those artifacts in public. Here's the exact way I'd run it as a beginner:

Step 1 — fork it, then do Phase 0 in one sitting

Fork the repo to YOUR GitHub — your fork becomes the public record of everything you build. Then knock out Phase 0 (12 setup lessons: terminal, Git, environment) in one sitting. Boring, essential, done.

Prefer a website to a repo? The official companion site aiengineeringfromscratch.com has the same 511 lessons with placement quizzes — also free, no signup.

Step 2 — pick ONE lane, not all 511 lessons

Nobody does a 330-hour curriculum linearly. Route by goal:

Build AI agents (fastest to impressive): Phase 0 → Phase 13 (Tools & Protocols — MCP servers, 31 lessons) → Phase 14 (Agent Engineering, 42 lessons). Roughly 100 hours to artifacts almost nobody your age has.
Understand LLMs for real: Phase 0 → 1 (Math) → 3 (Deep Learning) → 7 (Transformers) → 10 (LLMs from Scratch — you pre-train a 124M-parameter mini-GPT). The credibility route.
Employable this year: Phase 0 → 13 → 14 → 17 (Infrastructure & Production) → one Phase 19 capstone (there are 85) shipped and deployed. That's a resume section, not a bullet.

Step 3 — keep every artifact

Each lesson's outputs/ folder is portfolio material. Commit it, push it, pin the repo on your profile. "Completed an AI course" is a line recruiters skip — agents, a mini-GPT, and MCP servers they can click into is proof.

Stack it with the free student offers

The skills this teaches are exactly what's paying right now — OpenAI's WebMCP hackathon pays $3,500 per winning MCP-powered site, students get $100 in free Codex credits to build with, and ChatGPT Plus is free for 4 months. All three guides:

⚡ All the free guides → joshualarosa.ai →

Codex $100 · free Plus · WebMCP hackathon · 50+ more

Keeping it honest: the ~330 hours is the repo's own estimate — the 85 capstone projects add hundreds more. The Claude certification prep inside is free but not affiliated with Anthropic. And on the stat you'll see in the repo's README ("84% of students use AI, 18% feel prepared") — it's uncited; the sourced versions are ~9 in 10 students using AI (Digital Education Council, 27,000+ students) and about 1 in 4 recent grads feeling fully prepared to use it at work (Hult Workplace Intelligence). Everything here was verified against the live repo on August 27, 2026.

Keep reading

Get the rest of “AI Engineering From Scratch: The Free 511-Lesson GitHub Curriculum (Beginner Roadmap)” + the PDF — free.

Drop your email to unlock the full guide and download it.