Hearth

• 

  Demo of simple, local health queries for elderly citizens

Inspiration

Elderly populations around the world are aging at an unprecedented rate, and there are simply not enough caretakers to meet their needs. Increasingly cut off from the world and from information, just living day to day becomes a hazard. Additionally, over 5 thousand senior citizens die each year from improperly using their prescription medicine by accident, which is something that can be easily prevented.

What it does

Provides a locally running source of information and communication ability for senior Citizens. Designed for concerned family members to add to elderly family member's phone.

Functionality:

• Can communicate with trusted contacts to auto reply for health checks. Sometimes older folks don't see or hear notifications, which can lead to worried relatives over reacting. Hearth can provide health checks, such as current location, battery percentage, and steps taken recently to ensure that older family members are healthy.
• Computer Vision for medical use - Allows users with poor vision to scan medicine bottles to ensure they are taking the right prescription amounts to prevent overdose deaths.
• Chat interface for medical questions - users can ask medical questions and get a quick idea whether they should visit a doctor or not.

How we built it

We used a small, quantized language model running locally via llama.cpp on Android. The application connects to the Android SMS app locally to allow communication with concerned relatives. The medicine scanning abilities utilize the same Qwen3.5-2b local model, so no information gets sent to the cloud. The entire application is completely offline, enhancing trust for less AI literate citizens.

Challenges I ran into:

Getting a model small enough to run on average (not flagship) hardware while preserving factual output from the model. Additionally, computer vision with Qwen3.5-2b is a bit more demanding for RAM, which is a major constraint for mobile devices.

Accomplishments that I'm proud of:

A fully offline chat and visual interface that can be used in airplane mode. The SMS check-in digest — which turns Hearth from "a local Siri" into something genuinely novel.

What we learned

On-device models are just now crossing the threshold where this is possible on phones people actually own. And for this audience, trust matters more than capability — if they don't believe it's private, they won't use it.

What's next for Hearth

Better model quality as on-device hardware improves. A caregiver dashboard so family members can see check-in summaries. And expanding the vision features beyond pill bottles: receipts, mail, food labels, anything an elderly person might need read to them. Also, I plan to integrate voice models in the future to ensure people with little to no vision can easily interface with this application.

Built With

• lamma.rn
• qwen
• react-native