Graph RAG
What I built
Ontology-validated triples, entity resolution, graph persistence/community summaries, local/global queries and provenance.
Result
Local graph retrieval reached 56.7% accuracy versus 48.3% for hybrid retrieval on a selected synthetic corpus and 60 questions. Query-mode results are documented below.
Evaluation setup and selected results
Local graph retrieval reached 56.7% accuracy versus 48.3% for hybrid retrieval on 60 questions over a selected 128-document synthetic corpus. The comparison applies to local graph queries on this subset. Provenance records link retrieved information back to source spans.
Compare the evaluation evidence →
About the evidence
Based on source review and saved project artifacts. Recorded measurements apply to the stated model, dataset and configuration; they are not claims of a current production deployment.