Adaptive fitness and nutrition planning system that automates workout and dietary decisions using three AI/ML agents across an Android app and laptop compute node, synced via Firebase.
FuelForm models fitness as a dynamic system where a user's physical state evolves over time. The application uses three AI/ML agents working together across devices to deliver personalized, low-effort fitness optimization, reducing the cognitive burden of planning workouts and meals by continuously adapting recommendations based on user behavior, recovery, and progress.
The Android app collects user health data through Android Health Connect and runs a lightweight general AI model locally on-device for fast decisions and coordination. A connected laptop acts as the primary compute hub running three dedicated AI/ML agents: a Nutrition Agent that adapts calorie targets, macro emphasis, and hydration goals; a Fitness Agent that adjusts workout intensity, duration, exercise categories, and rest days; and a Chat Q&A Agent that answers health and fitness questions. Firebase serves as the backend and real-time synchronization layer.
Key features include sign-up/login for data security, initial self-reporting of preferences and health goals, a home screen with cat-meme-based adherence visualization, a chat interface for the Q&A agent, personalized nutrition plans accounting for dietary restrictions, weekly exercise plans split across workout types, a reminders screen with push notifications for meal/water/exercise logging, and a bonus donut finder feature showing nearby donut shops.
Built during the Snapdragon Hackathon at Columbia University with a team of 5. The project runs on Snapdragon S25 Elite hardware for on-device inference and includes dedicated unit tests and thorough documentation.
class AgentSync(private val firebase: FirebaseDatabase) {
private val nutritionAgent = NutritionAgent()
private val fitnessAgent = FitnessAgent()
private val chatAgent = ChatQAAgent()
suspend fun syncRecommendations(userId: String) {
val userState = firebase.getReference("users/$userId/state")
.get().await().getValue<UserState>()
val nutrition = nutritionAgent.adapt(
userState.dietaryRestrictions,
userState.calorieHistory,
userState.adherenceScore
)
val fitness = fitnessAgent.adapt(
userState.recoveryMetrics,
userState.workoutHistory,
userState.progressTrend
)
firebase.getReference("users/$userId/recommendations")
.setValue(Recommendations(nutrition, fitness))
}
}