523 lessons and every one of them ships something you can run
56,463 stars, still adding around 310 a day six months after it started, and the thing that separates ai-engineering-from-scratch from the pile of AI curriculum repos is one design decision: every lesson ends with a reusable artifact. Not a finished notebook you close and forget. A prompt, a skill, an agent, or an MCP server that goes into your toolkit and stays there.
The scope is not small. 523 lessons, 20 phases, roughly 342 hours, spanning Python, TypeScript, Rust and Julia. Phase 0 is dev environment setup, Phase 1 is linear algebra intuition, Phase 11 is LLM engineering, Phase 14 is agent engineering starting from the agent loop itself. There's a separate learning path for using coding agents on real repositories, which is the skill most people need and the one nobody teaches, because it's about workflow rather than model internals. MIT licensed, readable on GitHub or on aiengineeringfromscratch.com with the same lesson code either way.
The framing statistic they lead with is the honest one: 84% of students already use AI tools and only 18% feel prepared to use them professionally. That is a gap between exposure and competence, and it isn't fixed by more exposure. Traffic suggests people believe the pitch β 114,584 readers and 181,995 page views in a 30-day window as of late August.
Two caveats before you commit. The translated lesson pages are machine translations on a separate branch, so treat English as canonical if you're following anything technical. And a 523-lesson curriculum is a star-bookmark magnet; the star count is measuring intent, not completion. Pick one of the goal-based entry paths and start in the middle. That's what the table at the top is for.
https://github.com/rohitg00/ai-engineering-from-scratch
← Back to all articles
The scope is not small. 523 lessons, 20 phases, roughly 342 hours, spanning Python, TypeScript, Rust and Julia. Phase 0 is dev environment setup, Phase 1 is linear algebra intuition, Phase 11 is LLM engineering, Phase 14 is agent engineering starting from the agent loop itself. There's a separate learning path for using coding agents on real repositories, which is the skill most people need and the one nobody teaches, because it's about workflow rather than model internals. MIT licensed, readable on GitHub or on aiengineeringfromscratch.com with the same lesson code either way.
The framing statistic they lead with is the honest one: 84% of students already use AI tools and only 18% feel prepared to use them professionally. That is a gap between exposure and competence, and it isn't fixed by more exposure. Traffic suggests people believe the pitch β 114,584 readers and 181,995 page views in a 30-day window as of late August.
Two caveats before you commit. The translated lesson pages are machine translations on a separate branch, so treat English as canonical if you're following anything technical. And a 523-lesson curriculum is a star-bookmark magnet; the star count is measuring intent, not completion. Pick one of the goal-based entry paths and start in the middle. That's what the table at the top is for.
https://github.com/rohitg00/ai-engineering-from-scratch
Comments