🥗 Phi-3 Nutrition AI Coach

An intelligent nutrition coaching system that combines CatBoost machine learning models with Microsoft's Phi-3 fine-tuned language model to provide personalized meal planning and nutritional guidance.

🌟 Features

• Personalized Nutritional Target Prediction: Uses CatBoost regression models to predict optimal daily calories, protein, carbs, and fats based on user profile
• AI-Powered Meal Planning: Fine-tuned Phi-3 model generates contextual nutrition advice and meal recommendations
• 7-Day Meal Schedule Generation: Automatically creates balanced weekly meal plans aligned with user goals
• Multi-Goal Support: Tailored plans for fat burn, muscle gain, and maintenance goals
• Recipe Database Integration: Smart meal selection from a comprehensive recipe dataset

🏗️ Architecture

┌─────────────────┐
│  User Profile   │
│  (Age, Weight,  │
│   Goal, etc.)   │
└────────┬────────┘
         │
         ▼
┌─────────────────────┐
│  CatBoost Models    │
│  (Predict Targets)  │
└────────┬────────────┘
         │
         ├─► Calories Target
         ├─► Protein Target
         ├─► Carbs Target
         └─► Fats Target
         │
         ▼
┌─────────────────────┐
│  Phi-3 Fine-tuned   │
│  (Generate Advice)  │
└────────┬────────────┘
         │
         ▼
┌─────────────────────┐
│  Meal Planner       │
│  (Recipe Selection) │
└────────┬────────────┘
         │
         ▼
┌─────────────────────┐
│  7-Day Meal Plan    │
└─────────────────────┘

📋 Prerequisites

• Python 3.8 or higher
• CUDA-capable GPU (recommended for model inference)
• At least 8GB RAM (16GB recommended)
• ~10GB disk space for models and datasets

🚀 Installation

1. Clone the Repository

git clone https://github.com/yourusername/phi3-nutrition-ai.git
cd phi3-nutrition-ai

2. Install Dependencies

pip install -r requirements.txt

3. Download Required Models

The project uses:

• Microsoft Phi-3-mini-4k-instruct (downloaded automatically)
• Fine-tuned checkpoint (place in phi3_finetuned/ folder)
• CatBoost models (place in user model/ folder)

4. Prepare Dataset

Ensure you have the following CSV files in the dataset/ folder:

• recipes.csv - Recipe database with nutritional information
• GYM.csv - User fitness and health data for training

📁 Project Structure

phi3-nutrition-ai/
├── README.md
├── requirements.txt
├── .gitignore
├── Nutrition_phi3 model.ipynb    # Main training notebook
├── meal_generate.ipynb            # Meal generation script
├── dataset/
│   ├── recipes.csv
│   └── GYM.csv
├── user model/
│   ├── catboost_reco_calories.cbm
│   ├── catboost_reco_protein.cbm
│   ├── catboost_reco_carbs.cbm
│   ├── catboost_reco_fats.cbm
│   ├── preprocess.py
│   └── schema.json
├── phi3_finetuned/               # Fine-tuned adapter checkpoint
├── phi3_merged_standard/         # Merged model (auto-generated)
└── evaluation/
    ├── phi3_confusion_matrix.png
    ├── phi3_text_evaluation.png
    ├── meal_plan_evaluation.png
    └── rouge_bert_evaluation.png

💻 Usage

Quick Start with Jupyter Notebook

jupyter notebook meal_generate.ipynb

Python Script Usage

from meal_generate import NutritionAI

# Initialize the system
app = NutritionAI()

# Define user profile
user_profile = {
    "Gender": "Female",
    "Age": 30,
    "Weight_kg": 65,
    "Height_cm": 165,
    "Goal": "muscle_gain",  # Options: fat_burn, muscle_gain, maintenance
    "Activity_Level": "Moderate",
    "BMI": 23.9,
    "Blood_Pressure_Systolic": 120,
    "Blood_Pressure_Diastolic": 80,
    "Cholesterol_Level": 180,
    "Blood_Sugar_Level": 90,
    "Daily_Steps": 6000,
    "Sleep_Hours": 7,
    "Chronic_Disease": "None",
    "Allergies": "None",
    "Dietary_Habits": "Non-Vegetarian",
    "Exercise_Frequency": "3-4 times a week"
}

# Generate personalized meal plan
targets, ai_guidance, weekly_plan = app.generate_7_day_plan(user_profile)

print(f"Daily Targets: {targets}")
print(f"AI Guidance: {ai_guidance}")
print(f"Weekly Plan: {weekly_plan}")

🎯 How It Works

1. Nutritional Target Prediction (CatBoost)

Four separate CatBoost regression models predict optimal daily intake:

• Calories: Total daily energy needs
• Protein: Muscle maintenance/growth requirements
• Carbohydrates: Energy and performance needs
• Fats: Hormonal balance and satiety

Input features include:

• Demographics (age, gender, height, weight)
• Health metrics (BMI, blood pressure, cholesterol, blood sugar)
• Lifestyle factors (activity level, sleep, steps)
• Goals (fat burn, muscle gain, maintenance)

2. AI Guidance Generation (Phi-3)

The fine-tuned Phi-3 model generates:

• Personalized nutrition advice
• Meal timing recommendations
• Hydration and supplement guidance
• Exercise integration tips

The model is optimized with:

• 4-bit quantization for faster inference
• LoRA adapters for efficient fine-tuning
• Goal-specific prompting for relevant advice

3. Meal Plan Assembly

The system:

• Filters recipes based on user goals and targets
• Balances macros across breakfast, lunch, dinner, and snacks
• Ensures variety throughout the 7-day plan
• Provides per-meal and daily nutritional summaries

📊 Model Performance

The system includes comprehensive evaluation metrics:

• Text Generation Quality: ROUGE and BERT scores
• Meal Plan Accuracy: Nutritional target alignment
• Classification Performance: Confusion matrices for goal prediction

See evaluation visualizations in the evaluation/ folder.

🔧 Training Your Own Model

Fine-tune Phi-3

1. Prepare your training dataset with nutrition Q&A pairs
2. Open Nutrition_phi3 model.ipynb
3. Follow the step-by-step training process
4. The notebook includes:
   • Data preprocessing
   • LoRA configuration
   • Training with gradient accumulation
   • Evaluation and metric visualization

Train CatBoost Models

CatBoost models can be retrained with new user data:

• Update dataset/GYM.csv with additional user profiles
• Run the CatBoost training cells in the notebook
• Models are automatically saved to user model/ folder

🌐 API Integration (Optional)

You can expose this system as a REST API using FastAPI:

from fastapi import FastAPI
from meal_generate import NutritionAI

app = FastAPI()
nutrition_ai = NutritionAI()

@app.post("/generate-plan")
async def generate_plan(user_profile: dict):
    targets, guidance, plan = nutrition_ai.generate_7_day_plan(user_profile)
    return {
        "targets": targets,
        "guidance": guidance,
        "weekly_plan": plan
    }

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes:

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• Microsoft for the Phi-3 model
• Hugging Face for the Transformers library
• CatBoost team for the gradient boosting framework
• Recipe dataset contributors

📧 Contact

For questions or feedback, please open an issue on GitHub.

🔮 Future Enhancements

• Add support for dietary restrictions (vegan, keto, paleo)
• Implement real-time nutritional tracking
• Mobile app integration
• Multi-language support
• Community recipe sharing
• Integration with fitness trackers
• Meal prep optimization

---

Note: This is an AI-powered nutrition system for informational purposes. Always consult with healthcare professionals for medical advice.