Serg Masís | AI Track | Compass AI & Data Summit 2025
summary
Demonstrates a working hybrid RAG pipeline on a 1915 cookbook: OCR into structured JSON via BAML, chunking to embeddings loaded into Chroma DB, entities and relationships loaded into neo4j, and a discoverable MCP server (FastMCP) wrapping a graph+vector retriever. Agents built with Crew AI (researcher + critic) call the MCP tool; evaluation example shows a returned ingredient list missing parmesan = 80% match. Prototype runs end-to-end and supports scaling tool discovery to many hundreds.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:38Agentic AI with graph RAG
- 3:02AI agent components
- 5:10LLMs vs agent workflows
- 8:23Agentic AI patterns
- 13:17LLM and framework selection
- 15:43Knowledge base prep
- 20:25Data extraction for RAG
- 25:04MCP tools for agents
- 27:38CrewAI agents
- 30:49Agent evaluation trifecta
- 33:19Context engineering
- 35:49Future of agentic AI
- 36:35Q&A
related talks
01h 26mThinking in Graphs: The Cognitive Foundations of Agentic AI
01h 00mFrom Planning To Production-Ready RAG
46mFullstack AI: The Building Blocks for Production Ready AI-Powered Software
46mArchitecting Efficient Agents with Semantic Routing and Mixture-of-Tools
25mAgents + Data Streaming for Real-Time Enterprise Intelligence
36mVibing with Data -- Multi-Agent Data Modeling and Constructi...
32mIntelligent Data Platform : Powered by Semantic Layer, MCP & Agents
58mAI-Enhanced Big Data: Integrating Private LLMs and Vector Databases