← Theory library

RAG & Vector Search theory

Chunking, embeddings, hybrid retrieval, reranking, and grounded answers.

Practise in the Code Lab →

Easy

1. Why retrieval-augmented generation existsGround answers in your own documents instead of model memory.
2. Embeddings and vector similaritySemantic meaning as geometry in high-dimensional space.
3. Chunking strategies that decide your quality ceilingBad chunks cannot be rescued by a better model.
4. Vector indexes: HNSW, IVF, and ANN trade-offsApproximate search buys speed with a recall knob.
5. Hybrid search: BM25 plus vectorsKeywords catch what embeddings miss, and vice versa.
6. Reranking with cross-encodersRetrieve broadly, then reorder precisely.

Medium

Hard