Highly personalized AI-fitness coach and friend that completely understands your needs and adapts with your daily health data and schedule. Always willing to make adjustments to your plans.
Inspiration: I have been struggling with my workout and staying fit at home without needing to go to gym, especially during college semester. Many days I have to eat from outside as well. Some days I walk a lot, while some days it is almost none. Therefore, I wanted to make something where people easily speak with AI agent as a friend and immediately get the desired results. For people like me, it is important to know the kind of meals one can eat from outside which will have the least effect, therefore, I came up with the idea of cheat meal. Our goal is to make fitness fun and accessible to all of students.
What it does: Fittergem is an AI-powered fitness and nutrition coach that creates personalized workout, diet, and lifestyle plans based on your health data, calendar schedule, and preferences. It integrates with Google Fit/Apple Health and Google Calendar to fit seamlessly into your life, helping users stay consistent and motivated effortlessly. It gives users the ability to be honest with AI and allowing to customize their diet, workout and cheat meal plan by just prompting the AI. It also allows the users to put a picture of their meal and get recommendations and automatically get their workout plan to compensate for the meal. Cheat meals are recommended based on the most nearby food places of the user. The workout plan will be based on user's calendar schedule and the user can easily customize it even more using natural language.
How we built it: The frontend of the app was build using Flutter and the backend was written in Python. The backend was deployed on Railway server. FastAPI and flask were used for communication between frontend and backend. Oauth2 was used to setup calendar api and Firebase was used to make Google and Microsoft sign up. PostgreSQL was used to store all the information and plans along with chats of the user.
Challenges we ran into: One of the main challenges which I ran into was organizing the memory of the AI such that the user can update any of his plans and the Database should store it constantly and display it under the respective plan section. In order to do so it makes one table with various rows to store all the information of the user and the information keeps getting updated without having to send the AI the entire list. Another challenge was how to find all the body measurements in the most accurate way. I tried Deepface machine learning model, however it was trained on celeb data and was really inefficient for common people. In the end I decided to use Mediapipe, Insightface and Chatgpt to find the measurements.
Accomplishments that we're proud of: I am proud of the fact that the onboarding of the app is relatively smooth and it requires really less input as compared to various other apps as most of the information we get is from the image inputted. The app allows users to simply speak with AI and the AI uses structured output to adjust all the plans and even schedule your workout plan in your google calendar. You can input your food image and then AI will make the necessary changes in your plans after asking you. Once you give access to your health data then the AI will automatically read it when you come back to the app and it will again make the adjustments. Even when you are travelling, it will get your location and nearby food places and make a new cheat meal plan.
What we learned: This space is really crowded with lot of big fitness apps. Therefore, we would like our users to be raw and real with AI and the AI will do everything for you, with natural language prompts.
What's next for Fittergem: We plan to add more features, and publish the android version in less than a week and then the IOS version really soon.
Log in or sign up for Devpost to join the conversation.