SOS MedField

Offline, reliable guidance to help get you through emergencies when you are in an SOS environment.

Comment

Inspiration

We are from UCSC, surrounded by redwood forests where cell service can be unreliable or completely unavailable. The idea came from a real experience where we ran into someone overdosing on fentanyl during a hike and realized how scary it is to not know what to do in an emergency. Even when people want to help, panic, lack of training, and no internet connection can make it difficult to act quickly. In situations like CPR, overdose response, choking, bleeding, or wilderness injuries, the first few minutes matter a lot. We wanted to build something that could guide someone through those first critical moments, especially when help is far away.

What it does

SOS MedField is an offline medical assistant designed for emergencies in remote or low-connectivity environments. It helps users describe what is happening, interprets the situation, and shows the most relevant emergency guidance from a local medical knowledge base. The app can guide users through situations like CPR, overdose response, choking, severe bleeding, allergic reactions, hypothermia, and other field emergencies.

The app is designed to be hands-free and panic-proof. For example, in a CPR scenario, it can walk the user through positioning the person, placing their hands correctly, starting compressions, and continuing with a steady rhythm. It uses spoken instructions so the user does not have to keep looking down at their phone during an emergency.

How we built it

We built SOS MedField using bare React Native so we could have more control over native mobile features, Android setup, offline functionality, and text-to-speech. We used a mobile-first architecture because the app is meant to work in the field, not just as a website or online chatbot.

For the intelligence layer, we used Gemma 4 as an input-decoding and classification layer. Gemma does not generate the actual medical instructions from scratch. Instead, we compiled verified emergency medical information into structured JSON files. Gemma interprets what the user is saying, identifies the likely emergency, and helps decide which local protocol or information should be shown.

Challenges we ran into

One of the biggest challenges was that we kept pivoting our idea as we better understood what would actually be useful in a real emergency. At first, it was tempting to build a general AI medical chatbot, but we realized that was too broad and not safe enough for emergency situations. We shifted toward a more controlled system where AI helps understand the user, but the actual emergency guidance comes from curated offline protocols. Android development was also a major challenge. Since we used bare React Native, we had to deal with native setup issues, Java and Android environment problems, emulator configuration, Gradle dependency issues, and text-to-speech setup. We also had to make decisions quickly because this was a hackathon, so we had to balance a big long-term vision with what we could realistically build in 24 hours. Another challenge was designing the app for panic. Emergency instructions cannot be long, vague, or buried in paragraphs. We had to think about how to break information into small steps, when to pause, when to ask if the user is ready, and how to make the app useful even if the user is stressed.

Accomplishments that we're proud of

We are proud that SOS MedField is built to be almost fully offline. That was one of the most important goals from the start because the app is meant for places like forests, hiking trails, remote roads, and low-connectivity areas.

We are also proud of our hybrid approach. Instead of using AI to invent medical instructions, we use Gemma 4 to understand the user and route them to structured JSON protocols. This makes the app more intelligent than a static first-aid guide, but safer and more controlled than a general chatbot.

What we learned

We learned that building a medical assistant is not just about giving information. It is about giving the right information at the right time, in a way that someone can actually follow under pressure.

We also learned that for emergency medical use cases, AI should not be the source of truth. It is better to use AI as a language and routing layer while keeping the actual instructions grounded in curated, structured resources.

Technically, we learned a lot about bare React Native, Android native setup, local model integration, offline-first design, text-to-speech, structured JSON knowledge bases, and state-based emergency flows. We also learned how much polish matters in a demo: a guided flow with short spoken instructions can feel much more useful than a chatbot that gives one long response.

What's next for SOS MedField

Next, we want to expand SOS MedField with a larger verified offline medical knowledge base while keeping lookup times fast. We want to add more emergency protocols, including severe bleeding, burns, fractures, heat stroke, hypothermia, allergic reactions, poisoning, overdose response, and wilderness-specific injuries.

We also want to improve the hands-free experience with better voice input, spoken step-by-step instructions, timers, CPR metronome support, and multilingual support. In the future, SOS MedField could also support satellite messaging, emergency contact alerts, location sharing, personal medical profiles, and one-tap rescue messages.

Long term, we see SOS MedField as more than just a chatbot. We want it to become an offline emergency companion for hikers, campers, students, travelers, and remote communities where internet access and medical help are not always immediately available.

Built With

Share this project:

Updates