Revolutionizing Automation: Retrieval-Augmentation with LLMs
summary
Shows a working RFP automation app that parses uploaded PDFs/Word/Excel, splits text with overlapping chunks, embeds chunks into a vector store (ChromaDB/Pinecone/Elastic), retrieves top-k contexts and feeds them with the query into a generator to auto‑fill answers that users can refine and export; claims full end‑to‑end automation, large reduction in hallucinations and cost versus retraining, and reports GPT‑4 gave the best quality in tests. chunking + embeddings and map/reduce or refine pipelines are the core engineering choices.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:45What is RAG
- 3:11Limitations of general-purpose LLMs
- 4:36RAG in practice
- 7:31RFP processing with RAG
- 9:17Demo: RAG RFP app
- 12:07RAG pipeline architecture
- 14:12Document loading and splitting
- 18:56Text embedding and vectors
- 21:01Document retrieval techniques
- 22:06Answer generation
- 24:41RAG advantages
- 26:26Q&A: Hallucinations
- 27:50Q&A: Unstructured docs
- 28:56Q&A: LLM selection
- 30:42Q&A: RFP automation
related talks
57mLLMs gone wild
01h 00mLLMs gone wild
36mTechniques to avoid common LLM pitfalls
48mGoing Beyond Prompt Engineering
50mRetrieval Augmented Generation Beyond Toy Demos
01h 28mFrom RAG to Riches Building Enterprise Search Systems with LLMs
18mLLMs and GenAI: A Guide to Building Your Own RAG Model for Document Interaction
35mBoost LLMs with Event Streaming and Retrieval Augmented Generation