09 — RAG¶
Retrieval-augmented generation: embeddings, vector databases, retrieval, and grounded answers.
Scope¶
Cheat sheets for building RAG systems end-to-end — from documents to grounded answers. Raw LLM usage belongs to 08-LLM.
Cheat Sheets¶
| Sheet | Covers |
|---|---|
| rag.md | Complete RAG pipeline: loading → chunking → embedding → retrieval → generation |
| embeddings.md | Embedding models: OpenAI, sentence-transformers, dimensions, cost |
| chunking.md | Text splitting: recursive, semantic, code-aware, overlap |
| retrieval.md | Retrieval methods: similarity, MMR, hybrid, multi-query |
| reranking.md | Post-retrieval: cross-encoders, Cohere rerank, quality improvement |
| rag-evaluation.md | Evaluation: faithfulness, relevance, RAGAS, metrics |
| faiss.md | FAISS: local vector search, indexing algorithms, GPU |
| chroma.md | ChromaDB: developer-friendly, metadata filtering, persistence |
| qdrant.md | Qdrant: production vector DB, payload filtering, hybrid search |
| pinecone.md | Pinecone: managed, serverless, namespaces, zero-ops |
| weaviate.md | Weaviate: hybrid search (BM25 + vector), GraphQL, modules |
| milvus.md | Milvus: high-performance, GPU acceleration, Zilliz Cloud |
Contribute¶
- Template: CHEATSHEET_TEMPLATE.md
- Guide: CONTRIBUTING.md
- Planned: pgvector, Elasticsearch, advanced RAG patterns (HyDE, self-RAG)