ChatGPT or GenUI? Building SNCF Connect's Assistant of Tomorrow
summary
SNCF Connect & Tech describe integrating their offers into ChatGPT using the Apps SDK and MCP, and compare that to building a custom conversational UI with technologies like GenUI. They explain their implementation choices, stacks and pitfalls (RAG, NLP history from OUIBot to 2024), focusing on limiting hallucinations, protecting user data, and avoiding harmful booking behavior; they also discuss whether to share a MCP server between both approaches and open questions before enabling seamless ticket purchases in ChatGPT.
AI-written summary · about
chapters (16)
- 0:00SNCF Connect assistant
- 1:47History of assistants
- 3:14ChatGPT demo
- 5:24Salamanca prototype
- 6:07MCP architecture
- 12:49MCP security
- 16:22MCP SDK
- 23:01MCP vs Google UCP
- 26:32Web MCP tools
- 29:23Mobile app integration
- 30:47Building interfaces
- 33:16Salamanca architecture
- 36:06GenUI dynamic UI
- 37:53Conversational AI Q&A
- 41:23Strategy recommendations
- 42:48Future outlook
related talks
53mBuilding an App with ChatGPT
40mChatGPT Widget App from 0 to Production: A Real Story
05mChatGPT is poorly designed. So I fixed it
01h 16mChatGPT - and How we Can Benefit from it
51mChatGPT as a developer's personal servant
01h 09mChatGPT and OpenAI API solutions: 👑 My Favourite Implementations
25mBuilding the shopping journeys of the future - ChatGPT Apps, Adaptive Interfaces and more...
02h 00mChatGPT & Prompt Engineering