Advanced Retrieval-Augmented Generation Techniques
summary
Applies RAG to a medical-chatbot pipeline using a 170,000 patient-summaries plus 1.4M-abstracts PMC patients dataset, indexes them into an open-source vector database, and demonstrates gains from semantic chunking, hybrid BM25+vector search, LLM-based query rewriting, automatic similarity-based autocut filtering, reranking with a heavyweight scorer, and domain tuning (MedRoN-70B fine-tuned on ~50B medical tokens). A full open-source demo and GitHub prototype show end-to-end retrieval, rerank, and generation working on real benchmarks.
AI-written summary · about
chapters (6)
- 0:00Introduction and RAG basics
- 8:51Improving RAG pipelines
- 12:23Better indexing strategies
- 25:05Enhancing retrieval methods
- 33:37Optimizing generation quality
- 40:24Open-source RAG demo and resources
related talks
50mRetrieval Augmented Generation Beyond Toy Demos
44mRAG for a Medical Company: Technical and Product Challenges AI & Data
15mRAG Beyond Chatbots
48mAn Introduction to Retrieval Augmented Generation
25mAdvanced RAG Systems: Scaling Retrieval-Augmented Generation in Production
21mHow RAG is Redefining AI Capabilities
28mImproving responses in generative AI applications
51mDeveloping a Custom RAG System