GraphRAG
RAG on a knowledge graph instead of vector search alone, so it can answer questions that connect several entities.
The problem
Vector search returns passages that "look like" the question, but a question such as "which teams took part in project X?" needs information scattered across many documents to be joined together.
How it works
Documents are turned into a graph of entities and relations. At query time the system identifies the entities in the question and walks the neighbouring nodes to collect structured context before handing it to the LLM.
Architecture
documents → extract entities + relations → knowledge graph
question → find entities → walk the neighbouring graph → context → LLM