archive – Clara Zang
summary
A two‑year effort brings LLMs and a retrieval‑augmented generation stack into production at AT&T with a claimed company‑wide assistant that cuts agent search time (one anecdote reports reduction from 15 minutes to 50 seconds) and improves satisfaction; pipeline deploys early 2023 with nationwide launch, hybrid vector/keyword search, chunking, reranking, hallucination controls, access controls and multilingual workarounds, and uses RAG pipeline in production since 2023 validated by 200 Qs human evaluation and tradeoffs on latency, cost and model choice.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:27Keyword search to RAG
- 5:42RAG at AT&T
- 7:49RAG pipeline components
- 10:20Fine-tuning and prompt engineering
- 13:07Evaluating RAG accuracy
- 15:31Production challenges
- 17:59Scaling and multilingual needs
- 19:27Future beyond RAG
related talks
47mClarita: Clariant’s journey to accelerate growth with GenAI and RAG(AIM305)
48mUnlocking LLM Potential: From Exploration to Integration
15mUntangling Real Products and Hallucinations, Postmortem of a Phone Agent Sock Finder
39mLeveraging AI and Emerging Technologies for Competitive Advantage
01h 00mLLMs gone wild
28mRetrieval-Augmented AI Agents for Entity Matching at Scale
30mBuilding generative AI-powered products
21mHarnessing Technological Innovation & Accelerating Uptake