Askalon Sentinel

Sentinel is a real-time drone acoustic detection app powered by the Askalon audio-classification stack. Your phone becomes a passive listener that identifies the acoustic signature of drones from FPV quadcopters to Shahed and instantly shares alerts with a live shared map.

WHY SENTINEL

Drone threats are moving faster than radar coverage. Sentinel puts an acoustic sensor in every pocket, turning the crowd into a distributed early-warning network.

KEY FEATURES

• Real-time detection — on-device inference identifies drone signatures in under a second, no cloud upload of your audio required by default
• Live shared map — see every active detection from Sentinel peers in your area, color-coded by threat category
• Shahed geofence — automatic switch to the specialized Shahed detector when you are inside declared risk zones (Ukraine, Levant, and other regions)
• Live Activity + Lock Screen alerts — 3-tap labelling (Shahed / FPV / False positive) directly from the lock screen to improve the model
• Silent background monitoring — Sentinel keeps listening even when your phone is locked, and pushes a notification the moment the acoustic classifier fires
• Privacy-first — audio never leaves your device except when you explicitly submit a false-positive report for research

HOW IT WORKS

Your phone's microphone streams a compressed audio feature vector to the Sentinel detection engine. When the acoustic fingerprint of a drone matches, Sentinel:

1. Fires a local alert on your device
2. Broadcasts the anonymized event to nearby Sentinel peers via encrypted WebSocket
3. Optionally invites you to label the detection to feed the next model retrain

WHO IS IT FOR

• Civilians in conflict zones needing extra warning ahead of overhead threats
• Defense-adjacent professionals evaluating acoustic sensing as a complement to radar
• Journalists and researchers documenting drone activity
• The Askalon research team collecting labelled data to keep improving the classifier

BETA NOTICE

Sentinel is currently in beta. Expect occasional false positives and detection gaps at longer ranges. Every labelled event you report is used to retrain the model. Support: [email protected]