Working guides to how AI systems are built: agent harnesses and the primitives that configure them, embeddings and vector search behind retrieval, and the neural network fundamentals underneath both — with course materials from MIT, Stanford, Berkeley, and Princeton.
Each study pack bundles the best course notes from multiple universities around a specific topic, giving you a comprehensive baseline guide from trusted academic sources.
A model on its own only produces text — the harness is what lets it read files, run commands, and remember. This is how you configure one: the real files behind instructions, skills, tools, commands, hooks, memory, plugins, and scheduling, using Claude Code as the worked example.
Keyword search misses anything worded differently, and a model cannot answer from data it never saw. Embeddings and vector search fix both: how text becomes vectors, how similarity search works, and how vector databases and RAG are built on top — with runnable examples.
You changed a prompt and it seems better, but you cannot prove it and you will not notice when it breaks. This is how to build a suite that can: the config file, the graders, testing reliability instead of one lucky run, and a CI gate that catches regressions.
Modern AI rests on a handful of ideas that get named constantly and explained rarely. What they actually mean: how neural networks learn, what makes transformers work, and the optimization behind both — drawn from MIT, Stanford, Berkeley, and Princeton course materials.
Your RAG demo worked and production does not: answers are confidently wrong, and the passage that would have answered the question was never retrieved. This is how to fix that — measuring retrieval on its own, chunking, hybrid search, reranking, and the index settings that quietly cost you recall.