AI Engineering From Scratch
416-lesson open-source curriculum that derives every AI algorithm from raw math before a single framework gets imported.
How the 8.3 was reached
- Quality of output8/10
60.6k stars, 10.4k forks and an active MIT repo; 523 lessons across 20 phases. github.com
- Ease of use8/10
Single npx skills add command to start, with an interactive tutor via Claude or Cursor; runs locally. aiengineeringfromscratch.com
- Pricing value10/10
Free, MIT-licensed, runs on your own machine. aiengineeringfromscratch.com
- API and integration qualityn/a
Course, not software that runs in a workflow.
- Solves the problem it claims to solve7/10
60k stars show wide use and four learning paths are documented; the claim of building from raw math is not independently verified. github.com
8.3/10 is the mean of the 4 criteria that apply. Scored on the five-criteria rubric v1, 29 September 2026. How scoring works
Pricing
- All 416 lessons
- 20 structured phases
- Python, TypeScript, Rust, and Julia code
- MIT licensed
- No signup required, runs locally
Key Features
- 416 lessons across 20 phases spanning linear algebra through autonomous agent swarms
- Math-first pedagogy: every algorithm derived on paper before any framework appears
- Multi-language examples in Python, TypeScript, Rust, and Julia
- Covers backprop, tokenizer, attention mechanism, and agent loop built entirely by hand
- Browser-based progress tracker with local persistence
Pros & Cons
Pros
- Completely free — no paywall, no account, no upsell path
- Unusually rigorous bottom-up approach builds genuine intuition, not pattern-matching
- Multi-language support makes learned concepts portable across ecosystems
- MIT licensed and fully self-hosted — no platform dependency or lock-in
Cons
- Many lessons still marked planned — curriculum is visibly incomplete
- No video content, live instruction, or community forum
- Progress is browser-local only — clears if storage is wiped, no cross-device sync
- No feedback mechanism, grading, or certificate on completion
AI Engineering From Scratch is a rare free resource that refuses to skip the math — every algorithm gets derived before a framework is touched. It's well-suited to self-motivated engineers who want deep understanding over fast shipping. The incomplete lesson backlog and absence of any community or feedback layer are the main reasons to temper enthusiasm.
Try AI Engineering From Scratch →Added to scored.tools on
Competitors to AI Engineering From Scratch
Other tools in the coding category worth comparing.
Hands-On Large Language Models
7.5/10Illustrated guide to understanding and building Large Language Models with over 275 custom figures.
Build a Large Language Model (From Scratch)
7.3/10Comprehensive guide to understanding and building large language models from first principles.
AI Agents for Beginners
8.5/10Microsoft's 12-lesson course for learning to build AI agents from scratch
Fastai
8.6/10Simplifies training fast and accurate neural networks using modern best practices
Machine Learning Systems Book
8.8/10Free two-volume textbook on machine learning systems engineering and deployment at scale
More Articles Featuring AI Engineering From Scratch
Learn Harness Engineering Review 2026: Free Course For AI Agent Builders
A free, project-based course teaching the environment and verification scaffolding that makes Claude Code and Codex agents actually reliable in production.
reviewLLM Engineer Toolkit Review 2026: Worth the Hype?
Honest review of LLM Engineer Toolkit by Kalyan KS — a free LinkedIn-based collection of LLM engineering Q&A and interview prep content. Useful, but limited.
reviewAI Engineering From Scratch Review 2026: Honest Take
A free, math-first AI curriculum that derives every algorithm by hand before touching a framework. Rigorous but visibly unfinished.