Advanced Enterprise RAG Systems
summary
Argues that enterprise RAG is solved by three concrete moves: retrieval-augmented generation (embed company docs into a vector store, run a ranker, feed top facts to the generator), a confidence-first reward (called Mockingbird) that only answers when sure, and an independent hallucination evaluation model that scores factual consistency in milliseconds. Gives hard numbers (best models still hallucinate ~1.5% of queries), shows a failing consumer device demo (Rabbit) and a working Vectara deployment used for finance and support.
AI-written summary · about
chapters (8)
- 0:00The future of AI interaction
- 2:17Examples and demos of AI
- 15:13Understanding hallucinations
- 23:57Addressing AI reliability
- 24:39Hallucination evaluation
- 25:37Enterprise AI implementation
- 27:21Vectara platform capabilities
- 28:59Use cases and conclusion
related talks
42mTo RAG or not to RAG
22mThe Power of RAG at Scale: A Year of Explosive Growth and Observab...
34mAvoiding RAG Sprawl in the Enterprise
44mRAG for a Medical Company: Technical and Product Challenges AI & Data
46mOptimizing Retrieval-Augmented Generation (RAG) for eCommerce
31mEvaluating RAG Applications: From Retrieval to Response Quality
39mRAGs to Riches: How Our Content Affects Retrieval Augmented Generation
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions