Building MetroMind: Notes on Combining AI and Full-Stack Dashboards

MetroMind started from a simple frustration: most "smart city" demos I'd seen were either a pretty map with no real data behind it, or a dashboard full of numbers with no spatial context. I wanted both in one place — a digital-twin view of city infrastructure that a planner could actually look at, backed by live analytics and AI-driven traffic prediction.

Why a digital twin, not just a dashboard

A dashboard tells you what is happening. A digital twin tells you where. Traffic congestion at one intersection means something very different depending on what's around it — nearby infrastructure, alternate routes, time of day. So the first architectural decision was to treat the map view and the analytics panels as two coordinated but separate surfaces, rather than trying to cram everything into a single chart-heavy screen.

Where the AI layer actually sits

The traffic-prediction model doesn't try to be the star of the interface. It quietly produces a trend signal that the dashboard renders as a simple indicator — up, down, stable — rather than exposing raw probabilities or confidence intervals. Early versions showed more of the model's internals, and it made the UI feel like a debugging tool instead of a planning tool. Simplifying the output was one of the more valuable changes I made late in the build.

What I'd do differently

MetroMind currently runs on simulated traffic data rather than a live open-data feed — that's the most obvious next step, and it's on the Innovation Lab roadmap in spirit even if not listed by name yet. I'd also like to add historical comparison so the digital twin isn't just a snapshot of "now."

If you want to see it running, it's live at metromind-steel.vercel.app, and the full breakdown is on the MetroMind project page.