How this could work safely at scale

A production version would obtain consent before any AI processing, minimise and encrypt data, keep API keys on the server, apply strict retention periods, and provide human review and appeal routes. Provider and authority connections would use approved, documented integrations only. The prototype implements none of these live connections; it demonstrates the citizen-facing workflow with synthetic local data.