ML//RAG//GraphRAG

- Microsoft's approach: build a knowledge graph from documents, then query the graph alongside vector search

Microsoft's approach: build a knowledge graph from documents, then query the graph alongside vector search

Extracts entities and relationships, builds community summaries at multiple granularity levels.

Better for "global" questions requiring synthesis across many documents — where pure vector retrieval fails.

Open sourced (2024). One of the most popular RAG trends alongside ColBERT/ColPali.