My specialty
GraphRAG & Knowledge-Graph Agents
Vector RAG retrieves passages; it does not understand how facts connect. I build GraphRAG systems - knowledge graphs feeding LangGraph agents structured, relationship-aware context - for the questions plain retrieval cannot answer.
What's included
When vector RAG plateaus, graphs win
Scope
- Knowledge-graph modeling (Neo4j, FalkorDB, NetworkX)
- Entity & relation extraction pipelines
- GraphRAG retrieval feeding LLM agents
- LangGraph multi-agent orchestration & routing
How it works
Engagement
- Model your domain as a graph schema
- Build extraction & ingestion pipeline
- Wire graph-aware retrieval into agents
- Evaluate against a vector-RAG baseline
Proof
- Three-layer knowledge-graph pipeline (FalkorDB/Cypher, 149 entity types) on a production code-analysis platform
- 7 specialist agents each served a tailored graph slice - fewer tokens, better output
- This site's own code-graph analysis is graph-driven
FAQ
Common Questions
What is GraphRAG and when is it better than normal RAG?
GraphRAG retrieves from a knowledge graph instead of (or alongside) a vector store, so the model sees how entities relate - not just isolated text chunks. It shines on multi-hop questions, aggregation across documents, and domains where relationships carry the meaning.
Related