Retrieval Patterns

Retrieval-Augmented Generation (RAG) and its variants are the most common pattern for grounding LLM outputs in external knowledge. The industry treats this as a data pipeline problem.

It's not. It's an injection surface. Every document you retrieve crosses a trust boundary and enters an interpreter that can't distinguish instruction from data.

These patterns document how retrieval actually works, where the security assumptions fail, and what you should do about it.

Patterns in this category

Naive RAG Query Rewriting HyDE Multi-Query Retrieval Contextual Compression Parent-Child Chunking Recursive Retrieval Self-RAG Graph RAG Fusion Retrieval Agentic RAG CRAG (Corrective RAG)Semantic Caching