Adaptive RAG routes different query types to different retrieval strategies. Here's the taxonomy of strategies and how to classify queries.
Agentic RAG lets the LLM decide when, how, and how many times to retrieve. It's the natural evolution beyond one-shot retrieval.
Corrective RAG evaluates retrieval quality and recovers from bad retrievals with alternate strategies. Here's the pattern.
GraphRAG extracts entities and relationships from documents, builds a knowledge graph, and retrieves by graph traversal. Strong for corpus-wide, relationship-heavy queries.
Multi-hop questions require chaining multiple retrievals. Here's the pattern for answering questions that need information from multiple steps.
Self-RAG has the model decide when to retrieve, which documents to use, and reflect on its own output. Here's the framework.