Turn CCTV into
answers you can defend.
Attendance, behavioural analysis and task-completion verification from ordinary camera footage. A model reads the video; a rule set you can read off a page decides the verdict — so it says the same thing every time.
“Just ask the AI
if the task was
done correctly.”
It gives a different answer on different days, and when it says no, nobody can explain why to the person it is about. In a care setting that is not a rough edge — it is disqualifying.
Three questions, one engine
Who was in the room, how they seemed, and whether the task was done properly — from the same footage, on the same infrastructure.
From camera to verdict
Four steps. The model does one of them; the rest is ordinary code you can audit.
A camera, a YouTube link or a file becomes sampled frames and a contact sheet.
The model labels each window of video with the step it shows, and how sure it is.
A grammar checks the observed order against the workflow you defined.
Anything uncertain is queued for a person to accept, correct or reject.
Your infrastructure,
your data.
Point it at your own PostgreSQL — Neon, Cloud SQL, Supabase or self-hosted — and it creates its tables there. Keep media in your own bucket. Pull footage from your own cameras. Every connection is tested against the real service before it is saved.
See it on your own footage
Open the console, bring in a clip, and read the verdict in under a minute.