JAM.AI

Tired of asking your doctor, 'What can I eat?' or stressing over every food choice? Meet JAM.AI, your personal diabetic health assistant. Get instant nutritional advice and take control of your diet.

Comment

JAM.AI

Table Of Contents

Overviews

Jam.AI is your personal health assistant for managing diabetes. Using artificial intelligence, Jam.AI provides tailored guidance and support, helping you monitor blood sugar levels, offering nutritional advice, and sending medication reminders. With Jam.AI, managing your diabetes has never been easier.

How we built it

Jam.AI is based on the GPT 3.5 model. We tuned our own model by feeding it around 600 lines of Q&A data in JSONL format about different scenarios for diabetic patients. Each line contains a person's age, height, weight, glucose level, and the choice of food in question. Our front-end utilized bootstrap icons and our fantastic logo. For back-end, we used flask to make our app. JavaScript to POST input from the user to our model, GET to update the HTML, and POST the output back to the user.

Challenges we ran into

Initially, we chose text-bison@001 as our base model. Yet, the tuning failed more than 10 times. We tried changing the JSONL format many times, reducing the dataset, yet nothing worked. So, we changed to Gemini. While training Gemini, we noticed it was taking more than 40 minutes to train our data. We also lost a lot of time debugging some security issues. We ended up using GPT 3.5 because it had less training time and more readable documentation. During this process, one of our teammates finished the frontend but was stuck with implementing a secure API process using JavaScript. We realized JavaScript CAN'T keep our API key secure. We opted to use Flask in Python to make an app and import the model. We spent the next 10 hours with no sleep debugging our app. The app received the API call, but the model didn't receive any content from the user's input. At 4 AM, our chatbot was finally running, but we have another problem: How do we deploy it? We used Heroku to deploy our app right now.

Accomplishments that we're proud of

We tuned our first LLM. This was a big accomplishment, considering we didn't know anything about the tuning process prior to this hackathon. We connected Python with our front-end. This was our second-time using a flask ever. The three of us had minimal knowledge about JavaScript and Python, yet reading those hefty documentations paid off!

What we learned

Use GPT if you are a beginner. Be careful with the format when making the training data. Pay attention to file paths.

Project Link

https://github.com/JosephDavisC/Jam.AI

What's next for Jam.AI

We envision Jam.AI expanding its usage not just for diabetic patients but also for other patients suffering from cholesterol issues, Metabolic Syndrome, and other diet-related problems. For more efficient usage, we plan to make a feature where the user can make their own profile that includes all the health information, such as height, weight, and dietary problems. Jam will then make a meal plan according to each patient's specific needs. This personalized approach will allow for more targeted nutritional guidance, which can significantly improve management and outcomes for various health conditions.

We plan to collaborate with healthcare providers to integrate Jam.AI directly into clinical settings. This integration will ensure that the meal plans and health suggestions provided by Jam.AI align with medical treatments and advice given by doctors. By doing so, we aim to create a seamless bridge between medical advice and everyday health management, enhancing patient adherence and satisfaction.

Share this project:

Updates