No RAGrets
summary
Describes a working RAG pipeline built on an S3 corpus and MySQL metadata that starts with OpenAI embeddings and Postgres vectors but moves to AWS Bedrock knowledge base using Titan embeddings and the Claude Opus LLM. Documents are chunked, embedded into 1,000–3,000 dimension vectors, queried with approximate nearest neighbor search that takes on the order of seconds to a minute, and returned with cited source paragraphs. System proves reliable enough to plan replacing the main search, with occasional mismatched highlights.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:55What is RAG
- 3:40RAG use cases
- 5:06How RAG works
- 9:00Vector databases
- 12:31Sharper’s RAG project
- 15:39Live demo
- 18:48Q&A
related talks
01h 28mFrom RAG to Riches Building Enterprise Search Systems with LLMs
53mBuild RAG from Scratch
21mHow RAG is Redefining AI Capabilities
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
28mImproving responses in generative AI applications
34mImplementing an open-source RAG with OpenSearch
25mSet Up an Open Source RAG in 30 Minutes
59mThe Evolution of RAG