SmartCare.ai

Inspiration

We have noticed that many illnesses are often overlooked because people are uneducated or unaware of how slight changes in their health can affect their risk probability. After visiting medical professionals, patients obtain stacks of paperwork which includes medical jargon that is not easily understood. Because of this, they miss out on vital information that may heavily affect their health and be the reason whether they develop a chronic illness or not. For example, Diabetes Type 2 can occur because of a person being pre-diabetic who does not tend to any of their symptoms or does not realize that changes in diet, weight, physical activity can increase their risk of developing it. It is also hard to monitor lab results and other vitals without having to go through multiple medical records.

What it does

SmartCare.ai will allow patients to take control of their own health and see trends from every medical visit from changes in average heart rate, blood pressure, O2 stats, etc. They can upload any medical document containing results from recent visits, lab results, appointments, etc. From this, our Generative AI model generates a comprehensible health summary explaining medical notes along with warnings and recommendations that can be used to aid lifestyle choices. This also provides patients with organized tabs including visit history, upcoming appointments, and even prescription information. It saves you time, reduces the stress of managing complex medical data, and empowers you to take control of your well-being. With secure, smart analytics at its core, this tool is your partner in navigating health with confidence and clarity.

How we built it

Our application is a full-stack web app built with Next.js on both the front end and the back end, with a MongoDB database for data persistence. We chose Next.js because it simplifies server-side rendering (SSR) and offers built-in API routes, enabling us to keep the entire project under one framework. We trained Gemini to parse lab reports and to detect anomalies within data and observe overall trends. We also used Gemini to summarize lab reports and explain important information in a digestible format. We used PyCaret 3 to train a model around Kaggle datasets for diabetes. We hosted this model on HuggingFace Spaces to allow for convenient access. We use Synthea to create synthetic patient records in the commonly used C-CDA standard to show historical data analysis.. Finally, we host the app on Vercel.

Challenges we ran into

We struggled trying to find the scope of our project and finding data sets that displayed the necessary medical information we needed to create our platform. We landed on focusing on diabetes as its one of the most diagnosed illnesses in the US, and would serve as a good proof of concept. We also ran into issues with parsing our historical patient records since they were stored in a XML file that was formatted with the C-CDA standard. We converted it to a bare JSON and parsed it in the front end for a cohesive experience.

What we learned

We learned how to bridge issues in healthcare with generative AI and be able to create a platform which can be accessed by anyone and help educate more people about their personal health. An AI-driven personal health records app demonstrates that clinical notes and medical documents can be taken and generate a personalized insights and early warnings by detecting subtle health patterns, translating complex information into accessible language, and empowering users to manage their health more effectively.

What's next for SmartCare.ai

We would like to expand SmartCare.ai to be able to analyze and detect risk factors for other chronic illnesses in the U.S. In addition, we look to provide AI generated alternative medicine recommendations for people who are unable to afford in person-consults or are dealing with acute illnesses.