Untangling Real Products and Hallucinations: Postmortem of a Sock-Finding Call Agent
summary
A team built a voice-only phone agent that lets pharmacists order hard-to-pronounce medical hosiery by combining large language models with structured agent orchestration, full-text search, and real-time voice UX. The talk covers the hybrid search approach (full-text plus AI), designing specialized LLM agents, and production challenges of deploying a conversational telephone ordering system.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:11Voice agent start
- 3:59LLM early limits
- 5:0340 compression products
- 8:38Natural search with LLM
- 10:01Scaling to 20k products
- 12:32RAG and vector DB failures
- 13:36Algolia solution and synonyms
- 15:44Final architecture with 3 LLM
- 16:48Phone UX optimization
- 18:53Recent improvements
- 20:41Conclusion
related talks
15mUntangling Real Products and Hallucinations, Postmortem of a Phone Agent Sock Finder
21mTalking to others for you: Leveraging machines that speak the “human conversation” protocol
17mBuilding Effective Voice Agents
21marchive – Clara Zang
56mAI Agents That Don’t Suck: How to Build Ones That Actually Drive Business Value
18mFrom POC to Profit: Deploying Customer-Facing AI Agents That Actually Work
25mAI Orchestration: Getting Super Agents to Complete a Mission
01h 04mBuilding Expertise Into Generated Conversations