Designing and Building the Orbis AI Platform
Our journey began by asking AI to “act like a guard”, but context proved essential, for example – a passer-by on a public street is usually deemed harmless, yet the same person in a closed industrial site is a serious security breach.
The breakthrough came with site and camera profiling, which considers site access type, operating hours, and site protocols, allowing AI to reason situationally. Site categories further refined behaviour, for example – children in a forest are normal, but in a construction site signal danger – greatly improving risk distinction.
This framework also unlocked applications beyond security, including pest control, fly tipping, environmental monitoring, and waste analysis, transforming a false-alarm solution into a multi-purpose intelligence platform.
To scale Orbis AI efficiently, we built a multi-tiered model ensuring that lightweight models filter events accurately, while large language models provide higher-level reasoning, delivering flexible, accurate AI that operates through solar or battery-powered units over a 4G network.