How it works
AI flags the risk. People decide.
A deliberately human-in-the-loop pipeline — designed for fire-safety compliance evidence, not autonomous action.
The pipeline
From camera view to audit evidence
Connect existing cameras
Add your RTSP/IP camera views. Fixed angles covering the zones that matter; no new hardware.
Draw fire-safety zones
A safety officer marks fire exits, extinguisher points, hose reels, panel clearance, and restricted storage — once.
Continuous evaluation
The AI watches those zones. A person or object that persists where it should not becomes a candidate.
Advisory event + evidence
After temporal validation, an advisory event is raised with a snapshot and a plain-language rule explanation.
Human verification
A trained operator confirms it, or marks it a false alarm with a mandatory reason. Nothing closes itself.
Corrective action & report
Confirmed events get an owner, a due date, and closure with proof — all captured in a daily audit-ready report.
Boundaries
What the model does — and does not — do
Fire Watch is
- An assistive analytics layer on existing CCTV
- Advisory events, always human-verified
- False alarms recorded with a documented reason
- On-premise, role-based, no facial recognition
Fire Watch is not
- A fire alarm or certified detection system
- A smoke or flame detector (deferred until validated)
- Evacuation, suppression, or fire-panel control
- A replacement for procedures or safety personnel
See it on your own cameras
A structured pilot measures detection, false alarms, and closure on your site.