Building Production-Grade RAG Apps | TDE Podcast #12 - Jason Liu скачать в хорошем качестве

Building Production-Grade RAG Apps | TDE Podcast #12 - Jason Liu

Jason Liu is an ML consultant and educator trusted by engineers from OpenAI, Anthropic, Google, and Amazon. Get his free newsletter course here: https://improvingrag.com/ -- 📅 Upcoming events: https://lu.ma/tde Intro - 0:00 What is RAG? - 1:45 How to build RAG systems - 3:19 Inventory vs capability - 5:09 How to Build a Terrible RAG System Ignore latency - 6:55 Don't show intermediate results - 8:50 Don't display source docs - 10:05 Ignore churn - 14:52 Use a single (generic) index - 17:16 No need for custom UIs - 21:25 We should train an LLM from scratch - 23:07 Don't manually curate inventory - 24:50 Don't involve MLEs at data ingestion - 27:15 Treat all inbound inventory the same - 28:23 No need for data-specific ingestion - 29:54 Never ask data provider for clean data - 31:41 Never need to cluster inventory - 33:05 Focusing on local evals and not A/B test - 35:00 Q&A Does ChatGPT use RAG - 37:04 Managing acquisition and delivery as a consultant - 37:27 Best way to implement vector DB - 39:55 Future of RAG - 40:43 Handling complex tasks with RAG - 42:03 Handing long docs in RAG - 43:10 Tools for debugging - 44:10 Langgraph - 45:20 Graph DB - 46:14 Best practices for incremental development - 47:52