The Etsy Gifting Assistant: From Prototype to Production | Interrupt 26

Derrick Kondo from Etsy's GenAI Enablement team walks through how Etsy built and shipped a conversational gifting assistant using LangChain, LangGraph, and LangSmith in just six weeks. The agent helps shoppers find the perfect gift through collaborative, iterative recommendations across Etsy's unique, unstructured inventory. 0:00 Introduction 0:38 Inventory search challenges 1:22 Why build an agent for gifting 2:04 Beta release results 2:36 Engineering overview 2:52 Solution architecture with LangChain 3:27 Agent tools and middleware design 4:37 Engineering challenges and solutions 4:47 Stopping repeated tool calls (spin) 5:48 Fixing listing ID hallucinations 6:36 Memory management and the terminal UI debugger 7:36 Streaming architecture and socket passing 9:21 Evaluation methodology overview 9:46 Pass-K trajectory testing 10:21 LLM judge for outcome evaluation 11:09 Etsy Agents CLI and LangSmith integration 11:32 Batch simulation for dataset generation 12:06 LangSmith deployments for online and offline parity 13:34 Automated judge alignment 13:56 LangSmith for judgment stability 14:08 Deploying to production 14:17 Etsy Agents monorepo and automated CI/CD 14:51 Six weeks from start to beta 15:12 Summary and closing thoughts Extra resources: • Everything we shipped at Interrupt: https://www.langchain.com/blog/interr... • Meet LangSmith Engine: https://www.langchain.com/blog/introd... • LangSmith Platform: https://www.langchain.com/langsmith-p... • About LangChain: https://www.langchain.com/