Don't tell me RAG is easy
summary
A team builds a RAG-backed product and finds it’s a process, not a checkbox: they use keyword + vector hybrid search, embeddings and tools like DocLink and PGVector, but hit real failures — a 10,000-page scanned PDF that needs OCR, chunking and deduplication, frequent hallucinations from small models, and mismatched scores between lexical and semantic engines. They ship incremental releases, log every query/context/response, score changes to prove improvement, and stop when top user intents are reliably served rather than aim for perfection.
AI-written summary · about
chapters (15)
- 0:00RAG introduction
- 1:02What is an LLM
- 2:48How RAG works
- 4:12RAG as exam analogy
- 5:15RAG team roles
- 8:02User stories for RAG
- 9:46RAG search strategies
- 12:13Vector search challenges
- 13:36Problematic documents
- 15:40RAG failure scenarios
- 16:01LLM hallucinations
- 17:43RAG pipeline development
- 19:05RAG deployment monitoring
- 20:52RAG real-world lessons
- 24:07Q&A
related talks
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
19mImproving RAG Systems - Key Challenges and Solutions
21mHow RAG is Redefining AI Capabilities
27mWhy Your RAG is Failing , And How Open Source Can Fix It
19mRAG in 2025: State of the Art and the Road Forward
56mRAG: from dumb implementation to serious results
11mRAG Evaluation Is Broken! Here's Why (And How to Fix It)
01h 28mFrom RAG to Riches Building Enterprise Search Systems with LLMs