Hayvn

Inspiration

Nursing homes are often understaffed, leaving caregivers responsible for more residents than they can continuously monitor. Even dedicated staff can miss a critical moment when their attention is needed elsewhere.

For one of our team members, this problem became personal through a recent high school project delivering handmade cards to nursing-home residents. Those visits, alongside personal experiences and stories about gaps in care, inspired us to ask: How can we help caregivers notice when someone needs help, even when that person cannot ask for it?**

What it does

Hayvn is a privacy-conscious monitoring system that helps caregivers identify potential falls and concerning movement patterns.

It represents a person’s movement through a 33-point body skeleton, allowing movement analysis without requiring their appearance to be displayed. It combines computer vision with resident-specific context, helping distinguish everyday behaviour from activity that deserves attention.

For example, nighttime walking may carry different significance for a resident with dementia than for someone following their usual routine. Hayvn brings that context into the review process and surfaces alerts through a caregiver dashboard.

How we built it

We used MediaPipe to extract body landmarks and combined mathematical motion checks with ST-GCN-based movement analysis. Gemini adds resident-profile context to help interpret flagged activity, produce understandable explanations, and reach a verdict.

Tiger Data provides persistent storage for resident profiles, while our Docker-based services run on Vultr. We connected the deployed application to helpgranny.health, making the prototype accessible through a public website.

Challenges we ran into

One challenge was distinguishing potentially dangerous movement from ordinary actions, such as bending down or lying down intentionally. A single pose tells only part of the story; movement over time and resident context both matter.

We also worked through the practical challenges of connecting our detection services, database, and dashboard, then deploying them with Docker, domain routing, and HTTPS.

Accomplishments that we're proud of

We brought a personally meaningful idea into a working prototype that connects pose analysis, resident profiles, contextual interpretation, and a caregiver interface.

We’re particularly proud of making privacy and individual context central to the design. Our goal is to give caregivers a useful signal they can understand and act on.

What we learned

We learned that detecting unusual movement and understanding its significance are separate problems. Useful alerts require both technical analysis and knowledge of the person being monitored.

We also gained experience integrating machine-learning services, cloud infrastructure, persistent storage, and a frontend into one deployed application.

What's next for Hayvn

Our next step is to get Hayvn into the hands of more users and potential care partners so we can continue improving our baseline model.

With consented testing and feedback, we want to better understand normal movement across different people and environments, evaluate missed events and false alerts, and refine the system accordingly. More representative testing will help us make Hayvn more reliable and useful for the caregivers it is designed to support.