Relay

Inspiration

Riya’s mom was a caregiver for her grandparents, and she was responsible for taking them to doctor’s appointments, managing their medications, keeping track of their conditions, etc. It was a taxing job since she had to remember all of this information, and she would often bring a lot of reports & notes to every single doctor’s visit because there was so much information that she had to relay between each different medical provider. Making these kinds of medical visits easier was what first inspired us to create this app.

What it does

Relay is a centralized database for your health information, made easy to build and maintain through AI. Rather than asking users to manually organize scattered medical records, Relay lets you add, update, and query your health data conversationally by using a biometric health vitals scanner and by talking or typing to Nora, our AI health assistant. Nora uses your accumulated health history to help you figure out what kind of care you actually need. If you've had a history of skin conditions and come in describing a new issue, Nora is more likely to point you toward a dermatologist rather than general guesswork because she never diagnoses, but she helps direct you to the right specialist and helps you prepare the right questions to ask them. Relay also makes it easy to move information between patient and provider by sharing what matters, when it matters, without needing yet another portal login.

How we built it

We first created a detailed workflow diagram in Canva, making sure anything we wanted to do was accounted for. We didn't want to jump into this blind. As we created the workflow, one of our members created a mock-up in Android Studio, which made visualizing the project significantly easier. Then, we worked on creating the groundwork of the program in Visual Studio Code. After that, we integrated the AI models we wanted to use, and finally, we polished the UI.

Challenges we ran into

Integrating three separate AI/identity platforms into one coherent, secure flow was by far our biggest challenge. Getting ElevenLabs' voice agent talking to a private, authenticated backend (rather than exposing API keys client-side) took real trial and error, and keeping Gemini's retrieval scope properly restricted so Nora never actually gave medical advice required careful design, not just a working API call. On top of that, several of us were building and debugging in parallel across local machines, which meant re-linking work to our shared Supabase backend more than once as pieces came together. Getting everything talking to each other, securely and reliably, took far longer than getting any one piece working in isolation. We also had to figure out how to mitigate friction for physicians.

Accomplishments that we're proud of

Michael's favorite accomplishment was the workflow diagram. Just looking at it all zoomed out put a massive smile on his face. Riya's favorite accomplishment was integrating Presage. She was super proud of how it went. Andrew's favorite accomplishment was implementing gemini into the chatbot since it was super exciting to learn how gemini interacts with AI's and ElevenLabs Miranda’s favorite accomplishment was building the ElevenLabs AI agent and integrating the voices into the “Ask Nora” AI call feature that is central to Relay's purpose.

What we learned

We learned how to create workflows to represent what we want our app to do, we learned how to use Codex and other AI agents to streamline the programming process, we learned how to integrate APIs into our apps, we learned how to use and program with android studio, we learned how to decide on and define features that would best fit the app, and prioritize the features we want to add due to time.