Mount one camera
A standard indoor camera goes 5 to 7 feet up a wall, with the room in view. Nothing for residents to wear, charge or remember.
VIGIL learns how each resident normally walks, stands and moves, flags changes weeks before they become emergencies, and alerts caregivers the moment someone falls.
The VIGIL app, sample resident
A slightly slower walk. A little more time sitting. A restless night. Each change is too small to notice on its own, until it becomes a fall or a hospital visit.
older adults in the study showing walking speed tracks long-term survival.
Source: Studenski et al., JAMA 2011"No one was comparing how she lived this month to how she lived last month."
On a visit to India, my grandmother mentioned that her vision had been blurry for months. She thought it was just aging. When she was finally checked, her blood sugar was dangerously high.
Nobody was careless. The change was simply too gradual to see. I built VIGIL to see it.
Staff can only check on each resident every few hours. VIGIL watches the hours in between.
A standard indoor camera goes 5 to 7 feet up a wall, with the room in view. Nothing for residents to wear, charge or remember.
A small computer in the room, or VIGIL's server over an encrypted link, finds 17 body points several times a second. No video is recorded or saved; only measurements are kept.
The VIGIL app shows each resident's trends against their own normal, and sends an alert the moment something needs a person.
A slow, steady walker may be perfectly well. A fast walker who suddenly slows down is the signal. VIGIL learns each resident's usual range in about 7 days and flags what drifts from it.
Measures the angle of the torso, not height, so squatting, bending and lying down to rest don't set it off.
An alert when a resident gets out of bed between 10 p.m. and 6 a.m., and an earlier heads-up when they sit up. Many falls happen on the way to the bathroom.
Read from body points as the resident goes about their day.
Shared rooms and visitors are told apart by face, build, height, walk and clothing together. A caregiver names each person once, and every correction teaches it.
Plain-language highlights, how quickly alerts were answered, and one tap to share it with family. No charts to decode.
Resting heart rate, heart rate variability and sleep can sit alongside the camera's measures. Optional, never required.
The torso goes horizontal fast and stays down for 3 seconds, outside the bed or couch.
A notification from the VIGIL app with generic text only, plus an email to anyone who has alert emails on.
Their name and the time are recorded, so the team knows someone is on the way.
An unacknowledged fall is sent again after 5 minutes, then every 15, until someone responds.
The 30 seconds before the fall, as a moving stick figure. Never video.
So VIGIL is built to see movement, not people's lives.
Every frame is analysed as it arrives, on the device in the room or on VIGIL's server over an encrypted link, then discarded. No video is recorded or saved. Fall replays are stick figures.
Faces are one of several cues for telling people apart. VIGIL keeps numbers, not face photos, only for that camera, and deletes them with the person. It can be turned off for any camera.
Walking speed, posture and activity, in the facility's own database, plus one photo per person for caregivers.
Push notifications carry generic text only, such as "VIGIL: possible fall". No names, photos or health data.
The app reaches VIGIL over HTTPS or a WireGuard-encrypted link, each account sees only its own cameras, and every data request needs a key.
VIGIL reports what changed and what that pattern is linked to. Decisions stay with the care team.
Measured with a stopwatch, steps counted aloud, and staged falls onto a mattress, in a home where VIGIL runs around the clock.
Against steps counted by hand over a 60-second walk.
A fast drop that stays down 3 seconds alerts caregivers. Updated detector being retested now; results will be posted here.
Continuously since Sep 22, 2026, analysing several frames a second.
Walking speed is estimated from step rate, so it catches a slower step rate but not shorter steps. VIGIL is a prototype, not clinically validated. The next step is a supervised pilot with real residents.
Rushil Mohan is a student at The Overlake School in Bellevue, Washington. He designed and built VIGIL's camera pipeline, its cloud service and relay box, and the iPhone app after watching a family member's health change for months before anyone noticed.
VIGIL is now looking for its first assisted living pilot partners.
Free pilot installations for a small number of assisted living communities near Bellevue, WA. Setup is handled for you, and nothing changes for residents.
Or email hello@vigilhealth.me