Building an Institutional AI Platform from the Inside

Many institutions exploring AI are evaluating vendor platforms or waiting for their ERP to add features. A small data team at a midsized institution took a different path: building Aurora, an internal AI platform that connects directly to existing campus systems through a layered architecture designed to be maintained by the same people who already manage those systems. This poster presents the architecture patterns behind Aurora, from identity resolution and data service abstraction to multi-agent AI pipelines with human-in-the-loop controls. The focus is not on what one institution built, but on the transferable design decisions that make institutional AI feasible for teams without dedicated AI engineering staff.