RAG Pipelines Letting You Down? Fitch Group's Solution for Financial Documents
summary
Three production RAG applications handle hundreds of thousands of financial reports by combining custom chunking with embeddings stored in Azure AI Search and Elasticsearch, plus intent-based routing to narrow search over hundreds of millions of chunks. Pipeline adds query rewriting, reranking, context compression and post-retrieval agents that loop back for missing facts; guard rails and permission filtering enforce access. A five‑to‑six engineer team iterates eight‑plus months to raise user thumbs‑up from ~33% to >80% while moving latency from ~30s to under five minutes.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 4:33Baseline RAG
- 6:23RAG Use Cases
- 7:11Building a Production RAG System
- 25:10Demos
- 29:30Current Limitations of RAG
related talks
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
01h 16mForget RAG Pipelines—Build Production Ready Agents in 15 Mins
16mEmpower your organization with production-ready RAG apps on Azure
25mComposable RAG Pipelines With OPEA Highlighting GraphRAG
32mBuilding Performant RAG Applications for Production
59mThe Evolution of RAG
11mRAG Evaluation Is Broken! Here's Why (And How to Fix It)
27mBuilding a Modular RAG Pipeline in Python