Proxy-Pointer RAG: Solving Entity and Relationship Sprawl in Large Knowledge Graphs

Large enterprise knowledge graphs struggle with entity and relationship sprawl, making data ingestion and reconciliation computationally expensive. A novel approach called Proxy-Pointer RAG addresses this by treating a vector index as a structural index for semantic localization. The method uses a document's hierarchical structure to create pointers, so a vector search hit retrieves the entire, structurally intact document section for analysis. This allows a synthesizer LLM to see the full context, shifting the burden of entity reconciliation from the expensive knowledge graph to a faster and more accurate vector retrieval pipeline.