Smarter agents, fewer guesses.
she/her
Shreya Sharma is a Solution Architect at Databricks, where she helps organisations across Australia take AI from flashy demo to dependable production. She specialises in generative AI and agentic systems on the Databricks Data Intelligence Platform, spanning everything from retrieval and evaluation to governance, deployment, and Lakebase (serverless Postgres on the lakehouse). Before Databricks, she spent three and a half years at Deloitte building data engineering pipelines, doing analytics, and building AI-powered solutions, giving her a practitioner's view of what it actually takes to put data and AI to work. Shreya is happiest when an AI agent stops hallucinating and starts shipping.
You built an AI demo. It was magical. Then you tried to ship it, and the grief began: 1. Denial ("the demo worked") 2. Anger ("why is my RAG retrieving garbage?!") 3. Bargaining ("just one more prompt tweak…") 4. Depression ("$40k a month, and it leaked PII") 5. Testing ("turns out 'vibes' isn't a metric") 6. Reconstruction ("wait, I need a UI and a pipeline too?") 7. and finally Acceptance This talk is a guided tour through the seven stages of AI engineering grief - the messy data, hallucinations, eval-free guesswork, and 3am incidents. And every stage is really a data problem in disguise.
