Building Multi-Agent RAG Systems
summary
Argues that production-grade knowledge assistants need four concrete layers: high-quality parsing and retrieval (not naive 512-token chunking), agentic reasoning for query decomposition and tool use, richer output generation like full reports, and deployable workflows. Describes LlamaIndex products including LlamaParse which claims to cut hallucinations and already has 30k+ users, 50M+ pages processed. Shows concrete techniques: hierarchical indexing, multimodal RAG, graph RAG, and hybrid vector + SQL queries with SingleStore to combine semantic search and structured filters.
AI-written summary · about
chapters (8)
- 0:00Introduction to assistants
- 1:11Limitations of naive RAG
- 4:40Data processing and retrieval
- 8:23Advanced document parsing
- 13:16Advanced indexing techniques
- 18:07Hybrid search and platforms
- 22:37Agentic reasoning
- 27:24Frameworks and use cases
related talks
29mThe Future of AI Assistants
35mAdvanced RAG Techniques with LlamaIndex
20mEnhancing RAG with Multi-Agent Integration
16mEffective agent design patterns in production
31mAdvanced Enterprise RAG Systems
42mTo RAG or not to RAG
47mBuilding and Testing AI-Agent Powered LLM Applications: A Live Demonstration
15mRAG Beyond Chatbots