Agentic AI · Computer Vision · 911 Deployment Simulation
A Security Operations Center thought-experiment. A 911 call comes in — NLP speech recognition pulls the suspect’s attributes from the caller, and that object-recognition criteria goes out to a mesh of cameras at intersections, drones, police patrols in car, on bike or on-foot.
How it works
1. 911 Call & decode — Speech recognition grabs the details: color, clothing, direction. 2. Drop the suspect object — They land on a real street near the caller. 3. Object detection attributes distributed — criteria goes out to all sensor’s AI processing. 4. Task & resolve — When a sensor detects the object, central AI is notified and evaluates available police assets — the nearest units are tasked to that location — the chase is on.
What you’re seeing on the map
Every colored cone is a sensor’s field of view. Tap a sensor badge to turn a layer on/off.
Surveillance cams — public CCTV tagged in OpenStreetMap; 90°, ~50 m. Starts off.
Vehicle Dash Cams — 10 patrol cars, ~102 m.
Bike cams — 10 bike units, ~83 m.
Officer Bodyworn Camera / digital radio — 10 on foot, ~68 m.
Traffic cameras — real intersection view from OpenStreetMap; ~80 m. Starts off.
Drones — 3 in the air, 360° overhead, ~110 m, no roads needed.
🎤Acoustic mics — a 27-sensor mesh that triangulates gunshot location. Starts off.
How to drive it
📞 Answer 911 Call — press to hear the caller’s voice. ⚡ Start Scenario / ↻ Reset — reposition police assets. Sensor badges — click to toggle each layer (traffic cams and mics start off).
Real data under the hood
It’s all real geography. The streets, intersection cameras (hover for images) and public CCTV views come straight from OpenStreetMap (via Overpass API), drawn with MapLibre over a ~2.5 mi area around Boston Common.