What you are looking at

BeKavod watches over an elderly man living at home with dementia. Ordinary home cameras record short clips when he moves. A vision model running on a small computer in the house watches every clip and writes down what happened: did he sleep, eat, drink, use the toilet, walk steadily, get up at night, fall. Family members read this journal instead of watching hours of video.

This page shows 11 real events from September 2026, picked by hand, with the model's answers translated from Russian. Nothing here updates and nothing can be changed.

1. One card, one event. On the left is the clip — press play to watch it. On the right: the time, how serious it is, and what kind of event it was.
2. The tags are what the model concluded: posture, gait, sleep, food, toilet, fall. Below them are the facts it relied on: the route through the rooms, the camera, how he was recognised, how long he walked or stood.
3. “What the model was asked” opens the whole reasoning: each question was asked separately (is it him? did he fall? how does he walk?), with the model's answer, its own words and how long it took.
4. Filters above the list show only one kind of event — for example only sleep or only unsteady walking.

Normal

everyday life: sleep, meals, toilet

Attention

worth a look: unsteady walking, activity at night

Alarm

a fall or leaving the house — the family gets a notification

The model is not perfect, and this demo does not hide it: read its answers next to the video and judge for yourself. In the real journal the family marks every answer as right or wrong, and those marks are used to make the model better.