Advanced RAG Techniques with LlamaIndex
summary
LlamaIndex presents concrete fixes for RAG pipelines: an open source LlamaIndex Python framework with recent releases like composable memory for agents, a coming multi‑agent framework, Neo4j knowledge‑graph retrieval, and a launched mixture‑of‑agents module; a parser called LlamaParse that handles ~80 file formats; hierarchical indexing that summarizes tables for vectors; reranking that shrinks top‑1000 retrieval to top‑10; ~20 eval modules; advice that page‑based chunking often outperforms semantic chunking and that KV‑cache selection is the next hard problem.
AI-written summary · about
chapters (15)
- 0:00LlamaIndex and RAG intro
- 4:42What is RAG
- 6:32Naive RAG limits
- 9:20Better data for RAG
- 11:29Parsing for RAG accuracy
- 13:58LlamaParse for clean docs
- 15:02Hierarchical retrieval
- 17:08Multimodal indexing
- 18:35Query routing strategies
- 21:03Agentic RAG tools
- 24:32Observability and production considerations
- 26:18Custom RAG use cases
- 27:44Evaluating RAG performance
- 30:04RAG future trends
- 32:21Q&A
related talks
32mBuilding Multi-Agent RAG Systems
16mEffective agent design patterns in production
42mUsing LlamaIndex for building Generative AI apps
42mTake Your RAG To The Max!
01h 12mBuild RAG using Llama 3.1
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
29mThe Future of AI Assistants
01h 00mKeynote: Building and Productionizing RAG