What is Graph-RAG?
Graph-RAG is retrieval-augmented generation that retrieves over a knowledge graph, so answers can follow explicit relationships between entities instead of relying on text similarity.
Also called: graph retrieval-augmented generation, knowledge graph RAG
In more detail
Standard RAG embeds documents as chunks and retrieves whichever read most similarly to the question. Graph-RAG additionally models entities and the relationships between them, so a query can traverse structure — which amendment supersedes which clause, or which party an obligation binds.
Why it matters
Questions whose answers depend on structure rather than wording are exactly where flat retrieval fails, and where a confident wrong answer is most damaging — contracts, clinical records, regulatory obligations.
How iLeaf approaches it
iLeaf models the domain's entities and relationships before indexing anything, builds a graded question set from the client's real documents, and measures retrieval accuracy against it before generating answers on top. Every answer cites its source.