Interesting approach. How does Semantica represent provenance across entities and relations in GraphRAG? Is it aligned with something like PROV-O or custom lineage tracking?
python
store = GraphStoreWithProvenance(provenance=True)
manager.track_entity(
entity_id="person_123",
source_document="doc.pdf",
source_location="page 42",
source_quote="John Doe was born in 1980..."
)
Bottom line: Audit-grade provenance tracking for high-stakes GraphRAG where every decision must be traceable back to source material.
TL;DR: W3C PROV-O compliant, not custom - designed for domains where audit trails matter.
Hi HN We built Semantica after struggling to explain why GraphRAG systems produced certain answers and where facts came from in production.
The goal is to make semantics and provenance explicit and queryable, rather than hidden in prompts or embeddings.
It’s open source and used in real systems. Feedback and criticism are very welcome.
Interesting approach. How does Semantica represent provenance across entities and relations in GraphRAG? Is it aligned with something like PROV-O or custom lineage tracking?
Great question! Semantica uses W3C PROV-O compliant provenance - not custom lineage tracking.
Standards-Based: Full PROV-O implementation (prov:Entity, prov:Activity, prov:Agent, prov:wasDerivedFrom)
Complete GraphRAG Coverage: Document → Chunk → Entity → Relationship → Query → Response
Graph Store Integration: Automatic tracking for Neo4j, FalkorDB, Neptune, and upcoming Apache AGE
Enterprise Features: SHA-256 checksums, source quotes, page numbers, bridge axiom chains
python store = GraphStoreWithProvenance(provenance=True) manager.track_entity( entity_id="person_123", source_document="doc.pdf", source_location="page 42", source_quote="John Doe was born in 1980..." ) Bottom line: Audit-grade provenance tracking for high-stakes GraphRAG where every decision must be traceable back to source material.
TL;DR: W3C PROV-O compliant, not custom - designed for domains where audit trails matter.
Interesting product and direction.