Retrieval Augmented Generation Beyond Toy Demos
summary
Covers a full RAG pipeline from ingestion through retrieval, augmentation and generation with concrete implementations: Chroma vector store, OpenAI embeddings, LangChain/LlamaIndex/Semantic Kernel, and Google custom search as a black box. Recommends semantic chunking (1,000‑token chunks, 200‑token overlap) and describes agentic chunking and multivector retrieval to cut costs. Shows citation via model function calling, a fact‑checker loop and guardrails, a clinical chatboard for an Italian hospital, and a working knowledge‑graph + multi‑agent system that generates tests for undocumented codebases.
AI-written summary · about
chapters (17)
- 0:00🚀 Introduction: What is Retrieval Augmented Generation (RAG)?
- 4:00The Basics of RAG: Retrieval and Generation Steps
- 6:00Using Embeddings in RAG: Vector Stores and Similarity Metrics
- 8:30Basic RAG Pipeline Demo: From PDFs to Generated Answers
- 12:00Challenges in Real-World RAG: Handling Follow-Up Questions
- 15:00Mitigation Strategies in RAG: Chunking, Retrieval, and Generation
- 18:00Semantic Chunking vs. Agentic Chunking: Pros and Cons
- 21:30How to Improve RAG with External Retrieval Systems
- 24:00Handling Complex Documents and Data Beyond Text
- 26:00Advanced Techniques: Graph RAG and Multi-Agent RAG
- 30:00When to Split Documents: Chunking Strategies for Large Texts
- 33:00Dealing with Structured Data in RAG
- 36:00Real-World Case Studies: Codebases and Financial Reports
- 39:00Improving RAG Pipelines with Metadata and Filtering
- 42:00Multimodal RAG: Combining Text, Images, and Structured Data
- 45:00Handling Hallucinations and Ensuring Correctness in Responses
- 48:00Final Thoughts: Building a Robust RAG System for Enterprises
related talks
48mAn Introduction to Retrieval Augmented Generation
21mHow RAG is Redefining AI Capabilities
51mDeveloping a Custom RAG System
01h 00mLLMs gone wild
01h 14mAn Introduction to Retrieval Augmented Generation
25mAdvanced RAG Systems: Scaling Retrieval-Augmented Generation in Production
43mAdvanced Retrieval-Augmented Generation Techniques
15mRAG Beyond Chatbots