Arvo

Businesses with cameras already on the building — dealerships, warehouses, construction sites — tend to get either nothing after hours or a motion alert for every cat and bin bag, which trains people to ignore their own system. Arvo had to make existing cameras speak up only when the risk warrants it, without ripping anything out or moving footage off site.

• An on-site gatewayA small box on the site's own network reads the cameras' sub-streams and runs detection locally. What travels to Arvo is an event — what, where, when, how sure — not the video.
• Zones and hours per siteOwners draw the parts of a view that matter, and every site keeps its own opening hours and time zone, so 1am at a restaurant and 1am at a dealership are scored differently.
• Explainable risk scoringEvery detection is scored the same way — restricted zone, site closed, vehicle alongside, time on scene, detector confidence — and the reasons are kept with the number.
• Incidents, not notificationsRelated detections on one camera become a single incident with its own timeline: every event that fed it, the stills kept, who looked at it and what they decided.
• Human verification pathOn a monitored plan, high-risk incidents go to a monitoring provider's operators before anyone is woken, and the difference between the software's call and the human's is recorded.
• A demo that runs the real engineVisitors place a person in a generated scene and watch it scored by the same code that scores a live event — no real footage involved.