From a RAG Demo to a RAG System
Time and room for this session are published in October. when the timetable goes live.
Everyone's building agents right now and almost every agent that looks something up is doing RAG underneath. When that retrieval is wrong, the agent is confidently wrong. So this is a talk about the layer the whole agentic boom stands on.
Shipping a RAG demo that works is fine. You point it at some documents, ask a question, get a clean sourced answer, and it feels solved.
But a demo and a system aren't the same thing and the gap between them is four things production asks of a RAG pipeline that a dev demo never does.
This talk walks the RAG pipeline end to end ingest, chunk, embed, index, retrieve, generate and shows what each stage demands once real users, messy documents, and real permissions arrive: document quality, index staleness, access control, and the silent retrieval failures you can't catch without observability.
Along the way: why vector search alone isn't enough, and what actually separates a prototype from a system.
