Inspiration
As Jay's brother has had Type 1 Diabetes all his life, Jay described one of the many challenges his brother continues to face as he has to repeatedly go to the doctor if his basal ratio, or the many other factors needed to be adjusted. Further, he also has to keep track of the number of carbohydrates he eats.
What it does
IntelliDose is essentially a personalized control algorithm that determines whether the person is eating or exercising. If they are eating, Intellidose will provide an intelligent dose of insulin every 30 minutes if the system determines the person to be in the process of eating.
How we built it
We used an integration between the Dexcom API for the login page, HTML/CSS styling on the front end, with our Machine Learning Model on the backend. In order to train our LSTM neural network, we used both tensorflow and scikit learn.
Challenges we ran into
Many of our initial challenges revolved around integrating the Dexcom Login page into our website as it was a completely new API for us to work with. Further, many times our neural networks would keep giving atrocious results, so we kept on working towards optimizing both the error of dosage and classification.
Accomplishments that we're proud of
Being able to create a system, which can always accurately predict the amount of insulin intake within 1.73 units was very impressive. Further, we were completely successful in our efforts to completely integrate the dexcom API with our very own hosted website. Further, with our classification model, we were able to distinguish between "starting to eat" and "anaerobic" exercise
What we learned
Not only did we learn and gain more experience forming neural networks and front-end development, but we also really started to understand much of the current efforts in forming a closed-loop system as a cure for diabetes. By researching this system, we realize that although millions of people suffer from Type 1 Diabetes, developments in a cure are still just developments even with advancing technology.
What's next for IntelliDose
Precision is paramount when it comes to insulin administration. Our immediate goal is to further reduce the margin of error in insulin dosage and to refine our classification accuracy to exceed 99%. This is crucial because the consequences of excessive insulin can be life-threatening. We are committed to advancing IntelliDose to ensure it offers a reliable and safe solution for individuals managing Type 1 Diabetes.
Log in or sign up for Devpost to join the conversation.