The RAG Stack We Landed On After 37 Fails
summary
A concrete RAG stack emerges after 37 fails: prototype with LlamaIndex in Google Colab and BGE small embeddings from Hugging Face, switch to Quadrant as the vector DB for scale, use a closed-model re-ranker (cohere) or Nvidia reranker post-retrieval, test LLMs like GPT‑4 then serve production with Olama or Hugging Face TGI in Docker, trace with Langsmith/Arise Phoenix and evaluate with raggas; naive retrieval fails, reranking and Dockerized pipeline work and scale.
AI-written summary · about
chapters (7)
- 0:00Intro
- 0:18RAG Components
- 3:30Knowledge Base
- 4:04RAG Overview
- 9:19Building Blocks
- 11:09Swapping Components
- 16:34RAG Evaluation Framework
related talks
42mTake Your RAG To The Max!
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
53mBuild RAG from Scratch
41mFrom RAGs to riches: Build an AI document interrogation app in 30 mins
30mRAG: Tricks from the trenches.
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
28mNavigating Challenges and Enhancing Performance of LLM based Applications
56mRAG: from dumb implementation to serious results