Weather chase platform — multi-source operator console
Radar + lightning + ADS-B + drone + vehicle telemetry fused into one console so the operator never tabs between tools.
An operator chasing a storm was juggling 7+ tabs and apps. Critical decisions (deploy / pull back / re-route) were being made on stale or missed data.
One screen. Radar + EUMETSAT MTG-LI lightning + UDM PTZ feeds + drone telemetry + van GPS + Starlink link health. Each pane is a thin lens over the underlying data fabric (siphon → prism).
Lightning-risk scoring (regression on time-to-10km) and vision-based confirmation that a tracked cell looks the way the radar says it does.
All deploy/abort decisions. The AI surfaces confidence, never authority.
Operator workload measurably lower. Decision quality unchanged or better. Critically, the system degrades gracefully when feeds drop — operator never gets a blank screen.
First version tried to auto-classify storm severity and present a single recommendation. Operators stopped looking at the underlying data, then got bitten when the model was wrong. We pulled the recommendation; kept the score.
AI as scorer and surfacing aid; never as decider for safety-critical work.
- 01If your output drives a decision a human is liable for, the AI must support the decision, not make it.
- 02Always show the model's confidence and the underlying signal, not just the output.
- 03Field-ops UIs need to assume bad connectivity, not best-case.