Multiple Knowledge Graphs for Sophisticated RAG
summary
An AI-first SaaS models two Neo4j graphs: a tenant-specific context graph (live updates from Slack, Drive, docs, embeddings in a vector DB) and an external tenant-independent market graph that tracks ~1,700 categories of software. Initial per-tenant databases with Fabric proved too costly, so a single tenant-scoped DB plus a global DB is used with constraints, triggers and tenant-ID parameterization; runs on GCP, LLMs power logic but graphs cut LLM calls and provide repeatable answers, and selective isolation is available for sensitive customers.
AI-written summary · about
chapters (7)
- 0:00Introduction and company background
- 1:10Kimata’s procurement platform overview
- 3:57Event-centric architecture and AI agents
- 6:25Context graphs as application differentiator
- 7:48Internal vs external knowledge graphs
- 9:59Tenant isolation challenges and solutions
- 12:51Future of multi-tenancy and graph benefits
related talks
12mAgentic GraphRAG: Autonomous Knowledge Graph Construction and Adaptive Retrieval
42mMaking OSS AI Real with Knowledge Graphs
21mAgentic GraphRAG: Autonomous Knowledge Graph Construction and Adaptive Retrieval
39mEnhancing RAG with Neo4j Knowledge Graph
25mEnhancing Job Matching With Knowledge Graphs and RAG
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
30mAgentic RAG with Knowledge Graphs Explained
28mData-Empowered AI Assistants