Building Knowledge Base for Real Estate Using Gemma and Neo4j
summary
Built a queryable knowledge base for decades of real estate projects by extracting vectors from PDFs and feeding a RAG pipeline using Gemma initially with MongoDB then migrating to Neo4j. Highlights: Gemma + MongoDB initial prototype exposed document storage, brute force vector search and privacy risks. Migration to Neo4j for 10x–100x speedups yields far lower latency, smaller storage footprint and graph-based relevance. Also addresses token limits, per-project access controls and plans for batching, metadata cleaning and user feedback loops.
AI-written summary · about
chapters (9)
- 0:00Introduction and client background
- 8:22Real estate data challenges
- 13:02Why not train a model
- 15:09RAG approach explained
- 17:57MongoDB vector challenges
- 20:28Switching to Neo4j
- 22:12Performance comparison results
- 23:38Future enhancements planned
- 26:06Q&A and resources
related talks
32mGraphRAG: Powering Up LLMs with Knowledge Graphs
15mGraphRAG-Powered AI Agents: Transforming Corporate Metadata into Intelligent Knowledge
30mAgentic RAG with Knowledge Graphs Explained
01h 17mGraph Database and Retrieval Augmented Generation
01h 00mGraph Mastery to AI
27mAdaptive Knowledge Graphs for Dynamic RAG Systems
30mPerformant GraphRAG Improving Neo4j to Drive Autonomous AI ...
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs