Each department ran its own telemetry - traffic, drainage, lighting, waste, permits - and the operations room reconciled them on a whiteboard during incidents. The gap between knowing and seeing was measured in minutes, which is the wrong unit for an emergency.

We normalised every feed into one event schema on Google Cloud, so a pump, a camera and a permit application all arrive as comparable observations. On top of that sits a BigQuery model and a Next.js wall view designed to be read from six metres away.

The first live test came during a summer storm season. Departments that had never shared data were looking at the same timeline, and the average time from first signal to a coordinated response dropped by 54%.

Half the project was negotiation, not code. We ran joint workshops where each department mapped its own fields onto a shared vocabulary, and we shipped read-only access to that mapping weeks before the dashboard existed.