Standard vector-only RAG (`pgvector` / Pinecone) excels at finding semantically similar paragraphs, but struggles with multi-hop relational questions like "Which microservices owned by Team Alpha depend on the deprecated billing database?" Answering relational questions accurately requires combining vector embeddings with a Property Knowledge Graph (FalkorDB / Neo4j) and structured long-term agent memory.
If you want to hire a freelance Knowledge Graph, FalkorDB/Neo4j, and long-term AI memory programmer, I am Aldian Fazrihady.
GraphRAG & Dual-Memory Architectures I Build
1. Dual-Memory Agent Orchestration (Short-Term Context + Knowledge Graph)
I design dual-memory systems (The Dual-Memory Architecture: Short-Term Context & Knowledge Graphs) that fuse fast working-memory buffers with persistent Cypher graph entities and edges.
2. Asynchronous Fact Extraction via NATS & FalkorDB
To keep user-facing chat responses sub-second, I decouple LLM entity/relationship extraction into background NATS workers (Asynchronous Memory Curation with NATS and FalkorDB) that resolve entity coreferences and update temporal graph edges.