From RAG to Riches Building Enterprise Search Systems with LLMs
summary
Claims a production RAG pipeline is feasible: document parsing, semantic chunking with 10–30% overlap, OpenAI/Claude embeddings stored in a vector DB, and retrieval modes like multi‑vector, recursive retrieval and query‑transformation to close lexical gaps. The presenter demonstrates automated code generation and tests using Cursor with Claude 3.7 to create parsers, chunkers, embedder and vector store; the demo runs, installs missing packages and completes basic tests. Evaluation focuses on precision, recall, faithfulness, latency and cost tradeoffs.
AI-written summary · about
chapters (10)
- 0:00Enterprise search with LLMs
- 1:50RAG fundamentals and use cases
- 8:20Chunking for data prep
- 12:35Advanced RAG retrieval
- 25:53Self-curing and query routing
- 28:29Semantic and overlapping chunking
- 35:34RAG performance metrics
- 46:00Programming RAG with Cursor
- 1:00:18Building RAG end-to-end
- 1:27:07Q&A
related talks
48mGoing Beyond Prompt Engineering
41mFrom RAGs to riches: Build an AI document interrogation app in 30 mins
59mThe Evolution of RAG
23mNo RAGrets
34mImplementing an open-source RAG with OpenSearch
53mFrom RAG to Riches: Secrets of Data-Augmented LLMs
53mBuild RAG from Scratch
02h 47mFrom naive to advanced RAG: the complete guide