Deploying data analytics agents to production at Miro

Connect an AI agent directly to your warehouse and it answers well for a week, then starts to hallucinate. When Miro first pointed an agent at Snowflake, accuracy on a set of more than 1,000 evals came in at just over 40%. In this session, John Joyce talks with Ronald Angel, lead product manager for the data organization at Miro, about how his team layered metadata, curated skills, and governed semantic views to raise accuracy to more than 90%. Attendees learn how Miro certifies evals, runs them in its release pipeline and on production traces, and turns business-user feedback into better context.