Building AI features is easy to demo and hard to operate. The real challenge usually isn’t the prompt — it’s the data.

This hands-on workshop is for software engineers who are building AI-powered applications and discovering that model calls are only the visible tip of the stack. Behind every useful AI feature is a data engineering problem: getting the right data, preparing it for downstream use, keeping it fresh, retrieving it reliably, and knowing when the system is quietly going off the rails.

We’ll walk through the core data patterns behind modern AI applications, including ingestion, transformation, chunking, enrichment, retrieval, evaluation, and observability. Along the way, we’ll look at the failure modes that show up after the demo works: stale context, weak retrieval, bad metadata, silent data quality issues, and pipelines that no longer match the shape of the application.

The goal is to give software engineers a practical mental model for the data layer of AI systems — and a clearer understanding of how DataOps practices like testing, versioning, monitoring, and traceability help turn AI features into production systems.

By the end of the workshop, attendees will have a grounded framework for thinking about AI systems as data systems, and a better sense of how to design, debug, and operate them in the real world.