Skip to content

AI-powered emergency coordination

From emergency detection to coordinated action.

Take command of a simulated emergency. Assign responders, optimize routes, and see AEGIS in action.

Built for human-supervised emergency response

Interactive Simulation — No Real Emergency Dispatch

Scene · Building A
Simulated scene
05101520253005101520x (m)DEBRISBLOCKEDPROPOSED · REV 1C-0019.5, 16.5 mC-00216.0, 3.8 mC-00318.3, 12.8 mR-013.5, 3.0 mR-0227.0, 13.3 mCAM-01 · PERCEPTION
Perception · demo playback
Entity inspectorSim

Hover or focus a marker to inspect it.

Incident statusSim
Scene status
ACTIVE
Tracked entities
03
Available responders
02
Optimization
READY
Command authorization
REQUIRED

Your decisions. A changing scene.

You're in Command. What Happens Next?

Not a recording. A hands-on, private-to-your-browser command experience. No accounts, setup, or connected hardware.

  1. 01Assess a simulated emergency.
  2. 02Review casualty observations.
  3. 03Set human-controlled priorities.
  4. 04Generate responder recommendations.
  5. 05Approve and dispatch virtual responders.
  6. 06React to changing conditions.
Try the Simulation

Interactive Simulation — No Real Emergency Dispatch

Command Center / PreviewSynthetic · read-only
Synthetic incident grid — no real locations or devicesCENTRAL · OPENEAST · OPENC1C2C3C4ABWEST STAGING → EAST ACCESS / SYNTHETIC GRID
4 unconfirmed casualties2 virtual respondersNo assignments dispatched

01The problem

In an emergency, information is not enough.

Most monitoring systems stop at detection and visualization. The hard part is turning what is seen into who goes where — and knowing the task actually reached them.

  1. 01

    Fragmented situational awareness

    Responders must interpret incomplete information from cameras, field observations, and disconnected systems.

  2. 02

    Resource allocation under pressure

    Limited responders must be matched to casualties based on confirmed priorities, availability, capabilities, and travel constraints.

  3. 03

    Delayed coordination

    Recommendations are only useful when they translate into clear assignments, physical task notifications, and confirmed acknowledgments.

02The AEGIS solution

One system. A complete response loop.

Six stages connect what the camera sees to a responder physically confirming a task — with a human decision in the middle.

ObserveMapOptimizeApproveDispatchAcknowledgeAdapt

  1. Stage 01 · Observe

    Computer vision detects and tracks people through a webcam using YOLO and ByteTrack.

    Each frame from the webcam passes through YOLO person detection. ByteTrack keeps a persistent track ID for each person across frames, so the system follows people rather than re-detecting strangers.

    Camera output is evidence for the operator, never a medical assessment.

    InWebcam framesOutPerson tracks with pixel boxes
    Logitech webcamYOLOByteTrack

03Technology

Intelligence across the entire response pipeline.

Four layers, each with a narrow job and a clear interface to the next. Data rises through the stack; acknowledgments flow back down.

PERCEPTIONL1 · 5 componentsINTELLIGENCEL2 · 5 componentsCOMMANDL3 · 6 componentsPHYSICAL RESPONSEL4 · 4 components

Intelligence

L2 / 4

Holds scenario state, recommends assignments, plans routes.

A FastAPI service in Python owns the revisioned scenario state. Google OR-Tools proposes assignments from human-confirmed priorities; A* plans obstacle-aware routes; a structured LLM advisory can explain — never decide.

  • FastAPI
  • Python
  • Google OR-Tools
  • A* pathfinding
  • LLM advisory
Receives
Tracks + scene positions
Sends
Revisioned state over REST + WebSocket
Feedback
Acknowledgments from Physical

04Human in the loop

AI recommends. Humans decide. Responders act.

Try it: the optimizer's recommendation on the left has no controls. Only a commander, on the right, can approve, reject, or dispatch.

AI optimization recommendation

Illustrative example

Machine · advisory only

Revision
rev 3
Casualty
C-001
Priority
IMMEDIATE · confirmed by command
Responder
R-01 · Available
Recommended
R-01 → C-001
Route
Shown only when supplied by the optimizer

Rationale

  • Highest human-confirmed priority among unassigned casualties.
  • R-01 is available and has a feasible obstacle-aware route.
  • R-02 remains free for the next confirmed casualty.

Command authorization

Human · decision authority

  1. Review recommendation
  2. 2Approve rev 3
  3. 3Dispatch to responder device
  4. 4Physical acknowledgment

Dispatch unlocks only after an explicit approval of this exact revision.

Awaiting command decision — nothing has been dispatched.

Example only · nothing is sent

AEGIS keeps critical decisions accountable by placing human authorization at the center of every dispatch.

05Physical response

From digital intelligence to physical response.

A dispatched task doesn't end on a screen. It lights an LED on the responder's device, and the loop closes only when someone physically presses the button.

Responder device · R-01Interactive diagram
USB-C12×8 LED MATRIXWiFimoduleARDUINO UNO R4 WiFiresponder device · Grove shieldGROVE LEDOFFGROVE BUTTONIDLE
  1. 1Incoming task notification
  2. 2LED indication
  3. 3Pending acknowledgment
  4. 4Physical button press
  5. 5Acknowledgment received
  6. 6Dashboard status updated

Command center · task row

R-01 → C-001No task dispatched

Pending means the task was sent but no one has confirmed it. Acknowledged means the responder pressed the button — the dashboard labels whether it came from the device or a rehearsal control.

06About

Built at HackPrinceton Fall 2026.

AEGIS is a hackathon prototype exploring how practical and physical AI can support — not replace — emergency incident commanders. It connects computer vision, spatial mapping, optimization, and responder hardware into one auditable loop.

  • Simulation only

    Every scenario is simulated. No real patient data and no real dispatch.

  • Humans set priority

    Casualty priority is entered by a person. Computer vision never assigns it.

  • No automatic dispatch

    Approval binds to one exact revision; dispatch is a separate explicit action.

  • Evidence, not diagnosis

    Camera observations are shown literally, with confidence and age — never as medical findings.

Command center

See the entire response loop in action.

Explore how AEGIS transforms a simulated emergency scene into coordinated, human-approved responder assignments.