// the find
rohitg00/ai-engineering-from-scratch
Learn it. Build it. Ship it for others.
A structured AI/ML curriculum with 503 lessons across 20 phases, covering everything from linear algebra through autonomous agent swarms in Python, TypeScript, Rust, and Julia. The core pedagogy — implement an algorithm from raw math before touching any framework — is the right instinct. Aimed at developers who want to understand what's happening inside the models they're calling, not just chain API calls.
The BUILD IT → USE IT lesson split is pedagogically honest: you write backprop by hand before PyTorch shows up, and a BPE tokenizer before using HuggingFace. That sequencing actually works. Each lesson ships a concrete artifact (prompts, SKILL.md files, MCP server skeletons) rather than just exercises that die in a notebook. The phase dependency graph is explicit and accurate — skipping the math phases and then wondering why attention makes no sense is the learner's problem, not the curriculum's. The /find-your-level and /check-understanding agent skills for self-placement are a nice practical application of the material itself.
The lesson numbering has visible rot already — Phase 7 claims 14 lessons but lists 16; Phase 10 has lessons numbered 01–22 then jumps to 25 and 34. That's a sign content is being bolted on without being integrated, and a 503-lesson curriculum that's structurally inconsistent will confuse learners mid-path. The 'four languages' claim is overstated: Rust appears in one tokenizer lesson and Julia is confined to early math phases — the actual multi-language coverage is Python plus occasional guests. The outputs/ directories in the repo are mostly .gitkeep, so the '503 artifacts you'll own by the end' framing is marketing — you only get them by doing the work, which is fine, but the README implies otherwise. At 31k stars, there's real risk this is a link-aggregator hit rather than a battle-tested curriculum; no data on completion rates or learner outcomes is presented.