Case study: GoBearcats OS

The problem. An athletic department's fan data lives everywhere: ticketing, CRM, fundraising, email, survey tools. Every question - "who are our lapsed season ticket holders near campus?" - meant finding the right system, the right export, and the right person who knew both. The data existed. The access didn't.

The bet. Instead of buying another dashboard product, build a thin, AI-native layer over the data we already have: one place where staff can ask questions in plain English and get answers from live data.

How it's built.

• Snowflake is the source of truth - the department's fan data platform, joined on a common contact ID across ticketing, giving, and engagement systems.
• A secured API gateway (Supabase edge functions) brokers queries, so no credential ever lives client-side and every query is scoped.
• The app itself is built on Lovable - React and Tailwind, iterated conversationally. The agent writes most of the UI; I direct, review, and wire the data.
• AI sits on both ends: agents helped build it, and agents power the question-answering inside it.

What I learned.

1. The gateway is the product. The UI changes weekly; the security boundary can't. Investing in a clean, scoped query API first made every later iteration cheap and safe.
2. AI app builders are real for internal tools. For a department tool with a known user base, the Lovable iteration loop beats a traditional build by an order of magnitude. The catch - the bot owns the commits - is a fair trade for internal software.
3. Live beats perfect. The first version answered three questions badly. Shipping it created the demand - and the requirements - for the version that answers thirty questions well.

Status. In active use, with ingest sync and new data sources as the ongoing frontier. It also quietly powers other tools in the portfolio - the same gateway feeds the department's feedback and signage projects.