FitPilot

FitPilot is a modern, professional-grade metabolic tracking and biomechanical training assistant. It combines a FastAPI MongoDB Atlas backend with a native Android Kotlin mobile application.

---

Technology Stack

Mobile Frontend

• Framework: Native Android Application
• Language: Kotlin
• SDK Compatibility: Min SDK 23, Target SDK 30
• Camera integration: Jetpack CameraX (high performance, lifecycle-aware camera stream)
• Biomechanical AI: On-device TensorFlow Lite (MoveNet & PoseNet) running live coordinates extraction, joint angle mathematics, and rep counter completely client-side.

Backend Service

• Framework: FastAPI
• ASGI Server: Uvicorn
• Async MongoDB Client: Motor
• AI Engine: Google Gemini 2.5 Flash API (via raw HTTPS/HTTPX requests)
• Security: Cryptographic password hashing (Bcrypt) & JWT session tokens (PyJWT)

Database

• Engine: MongoDB Atlas (Async Document Store)

---

App Screens & User Flow

Here is a visual walkthrough of the application interface and user flow:

1. Secure Authentication	2. Daily Activity Dashboard	3. Indian Food Suggestions

4. Live AI Pose Tracker	5. Conversational Food AI Chatbot

---

Key Features

1. Daily Activity Dashboard

• Step Tracker: Accesses the native hardware pedometer. Tracks live step counts alongside calculated walk distance (km) and active calorie expenditure. Falls back to a mock step simulator on emulators/web.
• AI Adaptation Engine: Analyzes your logged meals and physical exercises for today, calculates metabolic targets based on body weight guidelines, and outputs macro and calorie guidance saved directly to your profile.
• Real-Time Performance Indicators: Dynamic grid highlighting active calorie burn, logged calories consumed, exercise exertion rate, and macronutrient balance (Protein/Carbs/Fat).

2. Conversational Meal Logger

• Natural Language Input: Type what you ate in plain text (e.g., "I had 2 cups of masoor dal and 4 boiled eggs for lunch").
• Gemini NLP Parsing: Translates plain text into ICMR standard portions, computes macro distributions, and auto-saves the logged meals straight into MongoDB.
• Persisted History: Loads previous chat conversation sessions on startup and supports clearing logs from the database via a trash icon.
• Custom List Formatting: Renders markdown details, bold labels, and bulleted ingredients into structured text bubbles.

3. Workout Logger

• Quick Logging: Save training sets, reps, and resistance weights.
• MET Calculations: Computes calorie burn and exertion intensity ratings dynamically based on exercise MET indexes:
  • Squat: 5.0 MET
  • Bicep Curl: 3.5 MET
• Workout History: Displays a timeline of exercises logged during the day.

4. Real-Time Posture Checker (CameraX & TFLite)

• On-Device Pose Estimation: Processes camera frames locally on the client using TensorFlow Lite MoveNet/PoseNet models via Jetpack CameraX.
• Form Analysis & Audio-Visual Feedback:
  • Squats: Checks depth and tracks alignment.
  • Bicep Curls: Checks elbow placement and counts reps.
  • Lunges & Presses: Dynamic pose calculation for multiple exercises.
  • Skeletal Color Cues: Skeleton overlays render in green when form is correct and turn red during errors.

---

System Equations & Calculations

Calorie Burn Calculation

Based on activity duration, metabolic index (MET), and user body weight:

$$\text{Duration (hours)} = \frac{\text{Sets}}{30}$$

$$\text{Calories Burned} = \text{MET} \times \text{User Weight (kg)} \times \text{Duration (hours)}$$

Exertion Intensity Score

Measures mechanical load adjusted against body weight:

$$\text{Intensity Score} = \left(\frac{\text{Sets} \times \text{Reps} \times \text{Weight (kg)}}{\text{User Weight (kg)}}\right) \times \text{MET}$$

Metabolic Goal Formulas

Calculated on user profile body weight:

• Target Protein: $1.6\text{g} \times \text{Weight (kg)}$
• Target Carbohydrates: $3.0\text{g} \times \text{Weight (kg)}$
• Target Lipids (Fat): $1.0\text{g} \times \text{Weight (kg)}$
• Calorie Allowance: $2000\text{ kcal} + \text{Workout Calories Burned}$

---

Getting Started

1. Setup MongoDB Atlas & Gemini API

Ensure you have:

1. A MongoDB Atlas Connection URI.
2. A Gemini API key from Google AI Studio.

Create a .env file in the backend/ directory:

SECRET_KEY=your_jwt_secret_signing_key
MONGODB_URL=mongodb+srv://<username>:<password>@cluster.mongodb.net/fitpilot
GEMINI_API_KEY=AIzaSy...

2. Launch the Backend Server

cd backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload --host 127.0.0.1 --port 8000

3. Build & Run the Mobile Application

You can compile and run the native Kotlin application using Android Studio or the Gradle wrapper command line:

cd FitPilotNative
./gradlew assembleDebug

The generated debug APK will be located at: FitPilotNative/app/build/outputs/apk/debug/app-debug.apk