P—09AI platform · SaaS + edgeIn development
Sentinel AI
AI CCTV intelligence where video never leaves the building.
- Type
- AI platform (SaaS + edge)
- Clients
- Web, edge node, mobile
- Automated tests
- 838 backend · 109 edge
- Technology
- LaravelReactPythonFastAPIONNXFlutter
Sentinel is an edge-first, cloud-optional security platform, still in active development. An edge node on site runs detection on camera streams; the cloud control plane only ever receives events — so sites with slow, expensive or unreliable connections still get real-time intelligence.
01Challenge
Expensive and unreliable bandwidth, frequent power interruptions, 8–64 cameras per site, and customers who will not accept a monthly bill measured in terabytes. The architecture had to start from the opposite assumption to the usual one.
02Approach
A Python edge node runs detection with a shared ONNX backbone and ByteTrack tracking, assembles events and spools them offline when the link drops. It pairs with the cloud by claim code and talks over an HMAC-signed API. The control plane covers sites and zones, an encrypted camera registry, recording timelines with gap detection, alert escalation ladders and a JSON rule engine with dry-run replay.
03Result
Twenty modules — from TOTP authentication and a hash-chained audit trail to visitor passes that verify offline, licence-plate recognition that never opens a barrier unattended, and face recognition that is off by default behind a consent ledger. 838 backend and 109 edge tests run without a model, GPU, camera or network.