ML//RAG//GraphRAG

GraphRAG is Microsoft's approach to retrieval: it builds a knowledge graph from documents, then queries the graph alongside vector search. It extracts entities and relationships, and builds community summaries at multiple levels of granularity.

GraphRAG is Microsoft's approach to retrieval: it builds a knowledge graph from documents, then queries the graph alongside vector search. It extracts entities and relationships, and builds community summaries at multiple levels of granularity.

It is better for global questions that require synthesis across many documents, where pure vector retrieval fails. It was open-sourced in 2024, and is one of the most popular RAG trends alongside ColBERT and ColPali.