Medical AI hackathon concept

AI-assisted caregiver monitoring

An extra set of eyes, without replacing yours.

Kinocular connects to a camera a family already has and watches for what a caregiver cannot always catch in person: a patient leaving the room unsupervised, a fall, or a repeated behavior such as head-banging. When something is flagged, the caregiver is notified right away, with enough detail to decide what to do next. Kinocular does not diagnose, and it never contacts emergency services on its own.

Watching only while the patient is alone.

The problem

A caregiver cannot watch every moment, and a camera alone will not tell them when to look.

Family caregivers looking after someone with a significant developmental, behavioral, or cognitive condition are often expected to provide supervision through most of the day, even though watching someone continuously is not realistic. They still need to sleep, work, cook, and step out of the room. A security camera or baby monitor does not close that gap by itself. It shows a live feed, but someone still has to be watching it to notice that anything is wrong.

Wandering, known clinically as elopement, shows how serious that gap can be. In a 2012 survey of 1,218 children with autism spectrum disorder, 49 percent had attempted to leave a safe space unsupervised at least once after age four, and 74 percent of those departures happened from the child's own home, the exact setting Kinocular is built to monitor.[1] Among the children who went missing, the average time away was 41.5 minutes: long enough for a traffic incident or drowning to occur, and both were common outcomes in the same survey.[1]

The consequences can be severe. An analysis of United States death records from 1999 to 2014 found that 27.9 percent of recorded deaths among individuals with autism were injury-related.[2] Self-injurious behavior is also far from rare: a population-based study of 8,065 children with autism spectrum disorder found a documented prevalence of 27.7 percent.[3] Kinocular is built around these two risks specifically: a patient leaving a monitored space, and visible self-injury such as repeated head-banging.

49%
of children with ASD surveyed had attempted to elope after age four
Anderson et al., 2012 [1]
74%
of elopements happened from the child's own home
Anderson et al., 2012 [1]
41.5 min
average time missing once a child had wandered off
Anderson et al., 2012 [1]
27.9%
of recorded deaths among individuals with autism were injury-related
Guan & Li, 2017 [2]
27.7%
average documented prevalence of self-injurious behavior in children with ASD
Soke et al., 2016 [3]

How it works

One continuous line of attention, from the camera to the caregiver's phone.

Kinocular is not built to understand everything happening in a room. It follows a specific, repeatable process designed to get one piece of information to the right person as quickly as possible.

Start with the camera already in the room

No proprietary hardware, sensors, or wearables. Kinocular runs as a software layer on top of a compatible camera feed that is already there.

Watch while the patient is alone

The system tracks presence and body movement for as long as the camera has coverage, without anyone needing to keep the feed open.

Recognize what matters

A departure from the room, a fall, prolonged stillness on the floor, or a predefined behavior like repeated head-banging is flagged the moment it is recognized.

Tell the caregiver

An alert reaches the caregiver's phone with the type of event, the time it happened, and a short description of why it was flagged. What happens next is their decision.

Keep a record

Every significant event joins a dated history, with the date, time, type, and description, so nothing depends on memory alone.

Notice change over time

Once enough history has built up, Kinocular can point out that a behavior such as head-banging has become more frequent, or has changed in a way worth a closer look.

What's built in

Every feature exists to shorten the time between an event and a caregiver knowing about it.

Awareness

What the system is watching for while the patient is alone.

Continuous monitoring

Tracks the patient's presence and movement for as long as the camera can see them.

Departure detection

Flags the moment a patient leaves the monitored room, since this is the single highest-risk event identified in the research.

Abnormal behavior detection

Recognizes a predefined set of concerning movement patterns, including falls, floor time, head-banging, and frantic movement.

Response

What happens the moment something is flagged.

Caregiver notifications

Sends what was seen, when, and why, then leaves the decision on what to do with the person who knows the patient best.

Multiple caregivers

Any number of trusted family members can monitor the same patient from their own device. One patient, unlimited caregivers.

Event logging

Keeps a dated record of every significant event, so patterns do not have to be reconstructed from memory.

Insight

What the system learns as more history builds up.

Pattern learning

Builds a picture of what is normal for this specific patient, so a one-off event is not treated the same as a real trend.

AI summaries and guidance

Surfaces frequency changes and recurring events in plain language, explaining what was observed without ever diagnosing it.

Safeguards

A monitoring product for people who cannot always consent to being watched needs firm limits.

These are not gaps in the roadmap. They are the boundary the product is designed around, agreed on before a single feature was built.

No facial recognition

Kinocular reads body movement and posture rather than faces, since recognizing how someone is moving does not require knowing who they are.

No medical diagnosis

It can flag a movement pattern as, for example, possible seizure-like movement or repeated head-banging, but it will never state that the patient is having a seizure. That judgment stays with caregivers and clinicians.

No automatic emergency calls

Kinocular notifies. It never contacts emergency services on its own. The caregiver always decides what happens next.

No proprietary hardware

It runs as a software layer on a camera the family already owns, rather than another device to install, charge, or replace.

The pause-and-notify rule

One person present
Monitoring active
Two or more present
Monitoring paused

Because facial recognition is out of scope, Kinocular has no reliable way to tell two people apart. The moment more than one person enters the frame, scanning pauses and caregivers are told monitoring is paused. Nothing recorded during that time enters the patient's history or pattern data, so the system never learns from, or mistakes, the wrong person for the patient. Monitoring resumes automatically the instant the patient is alone again, which is also when they are most at risk.

Evidence

The research behind the product.

These three studies establish the scale of wandering, injury, and self-injury risk for this population. They demonstrate why the problem is worth solving. They do not prove that any monitoring system prevents these outcomes.

  1. Anderson C, et al. Occurrence and Family Impact of Elopement in Children With Autism Spectrum Disorders. Pediatrics, 2012. pmc.ncbi.nlm.nih.gov/articles/PMC4524545
  2. Guan J, Li G. Injury Mortality in Individuals With Autism. American Journal of Public Health, 2017;107(5):791 to 793. pubmed.ncbi.nlm.nih.gov/28323463
  3. Soke GN, et al. Prevalence of Self-injurious Behaviors among Children with Autism Spectrum Disorder: A Population-Based Study. Journal of Autism and Developmental Disorders, 2016;46(11):3607 to 3614. pmc.ncbi.nlm.nih.gov/articles/PMC5392775

On the 39.89 figure sometimes cited alongside this research: that number is a proportionate mortality ratio for drowning, not a "times more likely" statistic. It reflects drowning's share of deaths in this population compared with the general public, and should not be quoted without that context.

Team

Built during a medical AI hackathon.

Ameen KhanHusam QutbMajed JiffriMohammed AlshawiSultan BakhshYousef Aqeel