Reduce AI Agents Costs with Semantic tool Selection
summary
Demonstrates a concrete fix: convert documents into a Neo4j knowledge graph and let the LLM generate text-to-Cypher queries so the database computes exact aggregates instead of the LLM guessing (example: a RAG estimate around 45 while the true count is 133). Builds two engines — a FAISS vector store retrieval tool and a Neo4j query tool — using Strands AI, OpenAI models and a local semantic transformer; demo with 300‑hotel data (light 10% option) shows GraphRAG returns precise counts and avoids fabrication.
AI-written summary · about
chapters (11)
- 0:00Welcome and Agenda
- 2:17Why RAG Hallucinates Numbers
- 3:18Graph RAG Explained
- 4:53Two Agents Demo Plan
- 5:49Building the Knowledge Graph
- 7:38Text to Cypher Queries
- 10:03Live Coding Setup
- 14:34Tools and Agent Loop
- 18:34RAG vs Graph RAG Tests
- 23:45When to Use Graph RAG
- 24:30Resources and Wrap Up
related talks
55mStop AI Agent Hallucinations: 5 Techniques + Production Patterns
15mGraphRAG-Powered AI Agents: Transforming Corporate Metadata into Intelligent Knowledge
21mAgentic GraphRAG: Autonomous Knowledge Graph Construction and Adaptive Retrieval
46mArchitecting Efficient Agents with Semantic Routing and Mixture-of-Tools
30mPerformant GraphRAG Improving Neo4j to Drive Autonomous AI ...
30mBuild an AI knowledge fabric for your organisation
29mAgentic GraphRAG: Multi-Agent Knowledge Graph Construction for Research Teams
34mJavaScript-Powered AI: Crafting Intelligent Systems with LLMs and Vector Databases