FuelForm is an adaptive fitness and nutrition planning system that automates workout and dietary decisions using AI and machine learning. The goal of the application is to reduce the cognitive burden of planning workouts and meals by continuously adapting recommendations based on user behavior, recovery, and progress.
FuelForm models fitness as a dynamic system where a user’s physical state evolves over time. The application uses three AI/ML models working together across devices to deliver personalized, low-effort fitness optimization.
FuelForm consists of an Android application and a laptop-based AI system connected through Firebase.
The Android app collects user health data through Android Health Connect and runs a lightweight general AI model locally to handle fast, on-device decisions and coordination. The phone syncs user state data to Firebase, which serves as the backend and communication layer.
A connected laptop acts as the primary compute hub and runs three dedicated AI/ML agents:
- A Nutrition Agent that adapts calorie targets, macro emphasis, and hydration goals
- A Fitness Agent that adapts workout intensity, duration, exercise category, and rest days
- A Chat Q/A Agent that answers any questions the user may have about their health and fitness goals.
Together, these models continuously adjust recommendations based on user activity, recovery, and adherence, minimizing manual planning while promoting sustainable long-term progress.
Note on Testing and Documentation: We have dedicated unit tests for our frontend and thorough documentation throughout the repository.
- Sign up/ Login Features for data security
- Preferences, health data, goals initial self reporting to use for fitness app
- Home screen that presents a silly cat meme (sad, indifferent, happy) based on how many reminders were addressed (with food, workouts, water)
- Chat screen to talk to the Chat Q/A agent (model hosted on phone) about health/fitnes
- Nutrition screen with a personalized nutrition plan based on dietary restrictions, personal data, and fitness goals
- Gym screen with personalized exercise plan for the week split into three different workouts
- Reminders screen to log the daily meals, water consumption, and exercises of the user (Sends notifications to the phone at the relevant times of the day to prompt the user to log these)
- Settings Screen to modify profile information
- Bonus Donut Finder Screen! We're confident our users will be on track with their diet and fitness goals. As such, we provided a donut finder feature to show the donut shops around wherever the user is at any time :)
-
Shruti Bhamidipati
Email: sb5197@columbia.edu -
Raghav Sampath
Email: rs4760@columbia.edu -
Kartik Kumar Gounder
Email: kkg2125@columbia.edu -
Anushk Pandey
Email: ap7151@nyu.edu -
Aditya Pendyala
Email: ap4839@columbia.edu
-
Android App (
fuelform.apk)- Runs a general AI model locally on the device to answer questions
- Displays workout and nutrition recommendations
- Provides helpful reminder notifications throughout the day
- Logs user meals and workouts daily
- Syncs user state with Firebase
-
Laptop (AI Compute Node)
- Runs the Nutrition AI/ML model
- Runs the Fitness AI/ML model
- Reads user state from Firebase
- Writes updated recommendations back to Firebase
-
Firebase Backend
- Stores user state and history
- Synchronizes data between the phone and laptop
- Enables real-time updates across devices
Android Phone
- Android 9 (API level 28) or higher
- USB debugging enabled (for APK installation)
Laptop
- Python 3.9 or higher
- Git
- Internet connection
-
Install the FuelForm APK on the Android device:
adb install fuelform.apk
-
Open the app and grant permissions for:
- Activity and exercise data
- Sleep data
- Nutrition and hydration data
- Body metrics (such as weight)
-
Ensure Android Health Connect is enabled and has granted FuelForm read access.
-
Clone the project repository:
git clone https://github.com/<your-org>/fuelform.git cd fuelform
-
Create and activate a Python virtual environment:
python3 -m venv venv source venv/bin/activate -
Install required dependencies:
pip install -r requirements.txt
-
Set up Firebase:
- Create a Firebase project
- Generate a Firebase service account key
- Save the key as
firebase_key.jsonin the project root
-
Export Firebase credentials:
export GOOGLE_APPLICATION_CREDENTIALS=firebase_key.json
-
Start the Fitness AI model:
python models/fitness_model/run.py
-
Start the Nutrition AI model:
python models/nutrition_model/run.py
Both models continuously listen for user state updates from Firebase and publish updated recommendations.
-
Launch the FuelForm app on the Android phone.
-
The app performs the following steps:
- Reads health data from Android Health Connect
- Runs the local general AI model for lightweight inference
- Syncs user state to Firebase
- Fetches updated workout and nutrition recommendations
-
Recommendations are displayed directly in the app UI.
- Open the FuelForm app daily.
- Allow the app to sync health data automatically.
- Review recommended workouts and nutrition targets.
- Follow the plan or log deviations as needed.
- The system adapts automatically over time based on adherence, recovery, and progress.
No manual planning is required from the user.
-
General AI Model (Android Device)
- Runs locally on the phone
- Handles lightweight inference and coordination
- Enables low-latency and offline-capable behavior
-
Nutrition AI Model (Laptop)
- Adjusts calorie targets, macro emphasis, and hydration
- Responds to training load and body composition trends
-
Fitness AI Model (Laptop)
- Selects training intensity, duration, exercise type, and rest days
- Adapts based on recovery, fatigue, and adherence patterns
FuelForm is a research and development prototype designed to demonstrate adaptive, AI-driven fitness planning using real-world health data and a multi-device architecture.
MIT License