I Want to Build a RAG System, Now What?
summary
Rapidly prototype a RAG POC: chunk documents into 400-token chunks, embed with an encoder, store in a vector DB (Pinecone/Canopy noted by uploader), retrieve top 15–20 results, run a simple ranker and feed that context to an LLM. Expect ~75% of specific queries to be answered by this baseline. Add planning/chain-of-thought steps or a single action loop to raise quality (one loop moves ~80%→95%). Log all queries, returned docs, prompts and scores to S3 for debugging and evaluation.
AI-written summary · about
chapters (8)
- 0:00RAG project introduction
- 6:31Data prep and POC
- 9:02Basic RAG architectures
- 13:14Advanced search and filtering
- 18:14Loop and iteration architecture
- 21:27Logs and debugging in RAG
- 22:57Data evolution and annotation
- 27:35Summary and final recommendations
related talks
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
56mRAG: from dumb implementation to serious results
42mTake Your RAG To The Max!
25mAdvanced RAG Systems: Scaling Retrieval-Augmented Generation in Production
53mBuild RAG from Scratch
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
02h 47mFrom naive to advanced RAG: the complete guide
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance