Folds duplicate reports into one incident and ranks every incident by urgency.
Speech-to-text transcribes each call live. The system extracts the type of emergency, the place named, identifying details (colors, vehicles, direction of travel), danger words in English and Spanish, and a voice-stress reading of how panicked the caller sounds.
Each report is scored against open incidents nearby: 45% distance (adjusted for location accuracy), 15% time, 25% emergency type, 15% shared details. ≥ 0.70 links automatically and can be split out in one click. 0.50–0.69 is kept separate and sent to a person for review. Different emergency types never merge.
Score bands: 70+ P1, 45+ P2, 25+ P3. Hard rules also apply: "not breathing", "trapped", "people inside" or a weapon always mean P1. A calm voice never lowers priority.
Every new report on a known incident is still read. If it adds danger, the whole incident moves up and the row shows which report raised it.
Incidents are stored in 1 km grid cells, so each report is compared only with nearby open incidents, and the work per report stays the same as call volume grows. Plugs into existing dispatch (CAD) systems and can exchange incidents in the NG911 standard format (NENA EIDO). If it fails, call-taking continues exactly as it does today.