Most people learn AI by calling APIs.
Very few learn what actually happens behind them.
I found a GitHub repository that flips this completely — AI Engineering From Scratch.
It's a free, open-source (MIT) curriculum that teaches you to build AI systems from first principles: 542 lessons across 20 phases, in Python, TypeScript, Rust, Julia and Go — with ~80k stars on GitHub.
Instead of just using frameworks, you build the internals yourself:
- Write backpropagation and a tokenizer by hand
- Build the attention mechanism and a transformer from scratch
- Implement retrieval pipelines and RAG systems
- Build agent runtimes, evaluations and MCP tooling
- Ship it all the way to production
Plus 104 projects — 79 ready to build — including a tiny coding agent, a browser agent, a multi-agent code review panel, document QA with citations, and an LLM gateway with fallbacks.
If you want to genuinely understand AI engineering instead of just prompting APIs, this is worth your time.
Explore it here: https://aiengineeringfromscratch.com
Repo: https://github.com/rohitg00/ai-engineering-from-scratch
Follow Anupam Mandal for more updates on AI and data engineering.
#AIEngineering #ArtificialIntelligence #MachineLearning #DataEngineering #LLM #OpenSource