SweetSpot

Inspiration

Yash, Yahli, Anirudh, and Vikram share multiple things in common, but none more than their passion for taking part in community initiatives. Everyone in the group possessed a passion for an accessible healthcare solution, specifically thinking about people who either were newly diagnosed or did not have an adequate support system.

We were thus motivated to build a resource with the idea that “you’re not alone”

Our idea to create a Type I Diabetes app stems from Vikram, who himself is a diabetic. The other three team members have watched in admiration as he diligently manages his sugars on a daily basis, and understand this is not necessarily the norm. Some people don’t have the resources or just don’t know how to take care of themselves. This product can fundamentally change that.

What it does

Our product is a mobile application that creates a personalized support center for diabetes management. The feature set consists of:

Dynamic Graph - View blood sugar level trends, updated with the latest data
Actionable Insights - Armed with tailored information, take action with our recommendation system to keep your sugars in check
Machine Learning Algorithm - Our model projects your sugars for the next 30 minutes, allowing you to stay a step ahead
Chatbot Service - How about some interactive advice? Converse with our chatbot to learn more about how to best manage your blood sugar levels based on your prior history

How we built it

The app is built on the foundation of the Dexcom API, which allowed us to pull all of our users sensor data and actionable events (such as exercise, giving insulin, etc.)

The Open AI API generates meaningful insights from diabetes monitoring data to help users understand their blood sugar patterns and adjust their habits for better health management. WHILE ALWAYS BEING SUPPORTIVE!

None of this of course would have been possible without the help of Firebase Functions - A serverless framework that let us automatically run backend code in response to event. This let us to make the connection between and backend seamlessly and make several API calls simultaneously or in sequence

We also experimented with a variety of machine learning/statistics techniques in order to ensure that our 30 minute glucose predictions were accurate. These are incredibly important because they are a key piece to our actionable insights and helping a T1D avert a low or high sugar long before it happens.

Finally, we created a Local LLM which is fined tuned. IMPORTANT! We can’t put P4 level data (HIPPA) into an online chatbot, so we pre-trained it to be able to analyze the users Dexcom data. It can then draw insights and finds patterns in users data. Finally, and perhaps most importantly, it can answer any diabetes related question, and also gives suggestions as to how to improve their overall blood sugar level numbers

Challenges we ran into

There were a number of challenges that we ran into throughout the course of our night. This list does not include everything, but gives a mere idea of what we faced.

Server crashing with multiple firebase functions running simultaneously
Figuring out how to create a trustworthy statistical model
Connecting the Neon Database to the front end client
Excess Memory usage of Google Cloud functionality

Accomplishments that we're proud of

First and foremost, we are proud of the fact that we built something that mattered to US. We could’ve opted to go down the route of building something with the sole intention of winning, but we instead felt motivated to build something that we could all get behind from the very core of what we stand for. The fact that one of our team members is Type I makes it even more special - we built it with his use case in mind. We’re also proud of the fact that we stuck it through. We stayed at the U Center the entirety of the 24 hours (to be fair, we did take our fair share of football/walk/social breaks) and it was honestly an unforgettable experience.

What we learned

We learned so much throughout this entire process. From teaching our not-incredibly-technical member how to work with databases and APIs, to building out a local LLM and using Firebase, there was a lot of technical knowledge accumulated over the past 24 hours. Additionally, we learned that we work really well together as a team in a technical setting. The 4 of us are all great friends, but we’ve never built a technical product together. Our technical skills (or lack thereof) really complemented each other and contributed to an environment defined by joyful collaboration and communal growth.

What's next for SweetSpot

We hope to iterate and pursue this as a product. As we mentioned, one of our team members is a Type I diabetic, so this is a cause near and dear to all of our hearts. We plan on refactoring our code base and then approaching Professor Sam King for advice on product-market fit and next steps. Not only is Professor King an incredibly accomplished entrepreneur, but he’s also a Type I Diabetic, so we hope to get both advisor and user perspectives from him.