Open-source 503-lesson AI Engineering roadmap: 20 phases, ~320 hours, MIT licensed
mdancho84 · x · 2026-09-09
A free, MIT-licensed AI engineering curriculum just went open source: 503 lessons across 20 phases (320 hours) covering Python, TypeScript, Rust and Julia. It runs from math/ML/deep learning foundations to Transformers, LLMs, RAG, MCP, agents, multi-agent systems and production AI. The distinctive teaching approach: build the underlying ideas first — backprop from the math, tokenizers, attention, LLM components and agent loops from scratch — then adopt production libraries, following a PROBLEM → CONCEPT → BUILD IT → USE IT structure.
Related event: Open-Source AI Engineer Curriculum Goes Viral with 54K Stars(3 posts)→
More from coding & agent
- Cognition's Devin agents factor RSA-260, setting a new public factoring record — jedisct1 · 2026-09-10
- How ML Intern Handles Compute Budgets: Safeguards That Don't Ruin the UX — Gradio · 2026-09-10
- Design Tools for LLMs, Not for Humans: Why Restricting Models Could Make Them More Capable — Revolutionary_Bed957 · 2026-09-10
- Running Deep Research on 100 Companies Costs ~$100: How Builders Are Cutting Agent Pipeline Bills — Confident-Green-5241 · 2026-09-10
- Hands-on: DeepSeek V4.1 Flash is fast at computer control, beats Codex for the author — AlchainHust · 2026-09-10
- Unity ships official Codex plugin with engineer-built skills for full game dev — lucasmeijer · 2026-09-10