A comprehensive web application that analyzes daily food intake and provides personalized nutrition insights using AI-powered recommendations and population health data.
- Natural Language Input: Enter meals in free text (e.g., "2 slices toast with butter")
- Smart Food Recognition: Advanced text parsing with 96.6% F1 score accuracy
- Comprehensive Nutrition Analysis: Calories, protein, carbs, fat breakdown
- AI-Powered Insights: Evidence-based recommendations using RAG (Retrieval Augmented Generation)
- Population Health Comparison: Compare your intake with NHANES demographic data
- Personalized Meal Recommendations: Get specific food suggestions based on nutrition gaps
- Clinical Notes Integration: Analyze medical nutrition recommendations
- AI Provider Comparison: Compare Gemini vs Ollama performance for research analysis
nutrition-analyzer/
├── app.py # Main Streamlit application
├── src/ # Core application modules
│ ├── clean_food_matcher.py # Primary food recognition engine
│ ├── llm_summarizer.py # AI summary generation with RAG
│ ├── meal_recommender.py # Personalized meal suggestions
│ ├── nhanes_analyzer.py # Population health analysis
│ ├── clinical_parser.py # Medical notes processing
│ └── rag_system.py # RAG implementation
├── data/ # Databases and guidelines
│ ├── nutrition_db_clean.csv # Curated 68-food database
│ ├── nhanes_demographics.csv # Population health data
│ └── nutrition_guidelines.txt # Evidence base for RAG
├── evaluation/ # Academic evaluation scripts
│ ├── training_script.py # USDA database training
│ ├── comprehensive_evaluation.py # System performance testing
│ ├── visualization_script.py # LLM comparison visualizations
│ └── overfitting_test.py # System validation
└── requirements.txt # Python dependencies
- Python 3.8+
- pip package manager
- Gemini API key (optional, has Ollama fallback)
-
Clone the repository:
git clone https://github.com/vinaysurtani/HealthAnalyzer.git cd nutrition-analyzer -
Install dependencies:
pip install -r requirements.txt
-
Set up environment (optional):
cp .env.example .env # Edit .env and add your Gemini API key for enhanced AI features -
Run the application:
streamlit run app.py
-
Open your browser to
http://localhost:8501
Input:
Breakfast: 2 slices whole wheat toast with butter and scrambled eggs
Lunch: Grilled chicken breast with brown rice and steamed broccoli
Snack: 1 apple with peanut butter
Dinner: Baked salmon with sweet potato and romaine lettuce salad
Output:
- Daily nutrition totals with target comparison
- Per-food breakdown table
- AI-generated evidence-based summary
- Personalized meal recommendations
- Population health comparison
- Database: 68 carefully curated foods with complete nutrition profiles
- Matching Algorithm: Multi-stage text processing with exact match priority
- Performance: 96.6% F1 score on comprehensive test suite
- Supported Foods: Common American/Western foods (bread, chicken, salmon, etc.)
- RAG System: FAISS vector database with SentenceTransformer embeddings
- LLM Integration: Gemini API with Ollama fallback
- Evidence Base: WHO, USDA guidelines, clinical nutrition research
- Provider Comparison: Gemini vs Ollama performance analysis
- Data Source: NHANES (National Health and Nutrition Examination Survey)
- Demographics: Age and gender-specific comparisons
- Sample Sizes: 1,000-1,500 people per demographic group
- Food Detection F1 Score: 96.6%
- Calorie Estimation Accuracy: 60%
- Processing Time: 0.12s (single food) to 0.67s (complex meals)
- Database Coverage: 68 foods across 5 major categories
- Gemini: 72.2/100 quality score, 6.29s response time
- Ollama: 66.0/100 quality score, 4.09s response time
- Both: 100% success rate, evidence-based responses
# Train USDA nutrition database system
python evaluation/training_script.py
# Comprehensive system evaluation
python evaluation/comprehensive_evaluation.py
# AI provider comparison with visualizations
export PYTHONPATH=/path/to/nutrition-analyzer && python evaluation/visualization_script.py
# System validation and overfitting analysis
python evaluation/overfitting_test.py- Add entries to
data/nutrition_db_clean.csv - Follow the format:
display_name,usda_name,calories_per_100g,protein_g,carbs_g,fat_g,serving_size_g,serving_description - Restart the application to reload the database
# .env file
GEMINI_API_KEY=your_gemini_api_key_here # Optional
STREAMLIT_SERVER_PORT=8501 # Default port- Target Calories: Adjustable in sidebar (1200-3000)
- Demographics: Age and gender for population comparison
- Clinical Integration: Paste medical nutrition notes
- AI Provider: Automatic Gemini → Ollama → Rule-based fallback
This system is designed for academic research in:
- Nutrition Informatics: Automated dietary assessment
- Population Health: Demographic nutrition analysis
- Clinical Decision Support: Medical nutrition therapy
- AI Comparison Studies: LLM provider performance analysis
- Human-Computer Interaction: Natural language nutrition interfaces
- File:
evaluation/training_script.py - Function: Trains USDA nutrition database system with TF-IDF matching
- Output: Trained model with 92.6% nutrition accuracy on 68 foods
- File:
evaluation/comprehensive_evaluation.py - Function: Tests CleanFoodMatcher system used in production
- Output: 96.6% F1 score, performance metrics, accuracy analysis
- File:
evaluation/visualization_script.py - Function: Compares Gemini vs Ollama AI providers with charts
- Output: Quality scores, response times, comparative visualizations
"No recognizable foods found"
- Use simpler food names (e.g., "chicken" instead of "grilled chicken breast")
- Check spelling of food items
- Refer to supported foods list in the app
"ModuleNotFoundError"
- Ensure all dependencies are installed:
pip install -r requirements.txt - Check Python version compatibility (3.8+)
Slow AI responses
- Check internet connection for Gemini API
- System automatically falls back to Ollama then rule-based responses
- Create issues for bugs or feature requests
- All evaluation scripts include detailed logging for debugging
This project is licensed under the MIT License - see the LICENSE file for details.