Gutty - AI Food Health Analysis

Direct ingredient identification and women's health analysis from food photos.

Takes a food image → identifies ingredients → provides health recommendations for PCOS, IBS, and Endometriosis.

Quick Start

# 1. Install
npm install
cp .env.example .env

# 2. Set Google Cloud credentials
export VERTEX_PROJECT_ID=your-project-id
export VERTEX_LOCATION=us-central1
export GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json

# 3. Build nutrition databases (one-time, ~30 seconds)
npx gutty nutrition-index
npx gutty gi-index  
npx gutty fodmap-index

# 4. Analyze food image
npx gutty gemini-health-analyze --image food.jpg --verticals pcos,ibs,endometriosis --out analysis.json

How It Works

Step 1: Gemini 2.5 Flash analyzes your food image and identifies individual ingredients Step 2: Ingredients are matched to nutrition database (COFID) for calories, macros, GI values Step 3: Rule-based health analysis generates warnings/recommendations for selected conditions

Total time: ~25 seconds per image

Prerequisites

1. Node.js 18+
2. Google Cloud Project with Vertex AI API enabled
3. Service account with Vertex AI permissions + JSON key file

Google Cloud Setup

# Create project, enable Vertex AI API, create service account with "Vertex AI User" role
# Download service account JSON, then:
export VERTEX_PROJECT_ID=your-project-id
export VERTEX_LOCATION=us-central1
export GOOGLE_APPLICATION_CREDENTIALS=/path/to/your-service-account.json

Database Setup (Required)

The system needs 3 local nutrition databases. Run once:

npx gutty nutrition-index     # UK COFID nutrition data (3,000 foods)
npx gutty gi-index           # Glycemic index values (2,000 foods)  
npx gutty fodmap-index       # FODMAP classifications (500 foods)

Expected output:

✓ Indexed 2847 nutrition records
✓ Indexed 1876 glycemic index records  
✓ Indexed 492 FODMAP records

Verify setup:

npx gutty validate           # Check API access and databases
ls ./lancedb/               # Should show: nutrition_data.lance, glycemic_index.lance, fodmap_data.lance

Usage

Main Command

npx gutty gemini-health-analyze --image PHOTO --verticals CONDITIONS --out OUTPUT.json

Parameters:

• --image: Path to food photo (JPG/PNG) or image URL
• --verticals: Health conditions to analyze (comma-separated): pcos, ibs, endometriosis
• --out: Output JSON file path (optional)

Example

npx gutty gemini-health-analyze --image curry.jpg --verticals pcos,ibs --out curry-analysis.json

Output Format

Complete structured JSON report:

{
  "step1_ingredient_identification": {
    "method": "gemini-2.5-flash",
    "identified_ingredients": [
      {"name": "Okra", "confidence": 0.9, "description": "Green cooked vegetable pieces"}
    ]
  },
  "step2_nutrition_mapping": {
    "ingredient_mapping": [
      {
        "ingredient": "Okra",
        "cofid_match": "Okra, boiled in unsalted water", 
        "glycemic_index": 54,
        "nutrition_per_100g": {"calories": 16, "protein_g": 1.8, "carbs_g": 1.9},
        "health_flags": ["low-gi", "pcos-friendly", "low-fodmap"]
      }
    ],
    "totals_calculated": {"calories": 131, "gi_value": 54, "is_low_gi": true}
  },
  "step3_health_analysis": {
    "analysis_by_vertical": {
      "pcos": {
        "triggers_found": [
          {"message": "Low GI (54) supports insulin sensitivity", "severity": "positive"}
        ]
      },
      "ibs": {
        "triggers_found": [
          {"message": "Low-FODMAP ingredients well-tolerated", "severity": "positive"}
        ]
      }
    }
  }
}

Health Analysis

PCOS/PCOD

• Analyzes: Glycemic index impact on insulin sensitivity
• Flags: High GI foods (>70), refined carbs, low fiber
• Recommends: Low GI alternatives, high fiber foods

IBS (Irritable Bowel Syndrome)

• Analyzes: FODMAP content based on Monash University research
• Flags: High FODMAP triggers (onions, garlic, wheat, certain fruits)
• Recommends: Low FODMAP alternatives, portion guidance

Endometriosis

• Analyzes: Anti-inflammatory potential, fiber content
• Flags: Processed foods, low fiber meals
• Recommends: Anti-inflammatory foods, fiber-rich options

Troubleshooting

Authentication Issues

# Check environment variables
echo $VERTEX_PROJECT_ID
echo $GOOGLE_APPLICATION_CREDENTIALS

# Test API access  
npx gutty validate

Database Issues

# Rebuild if corrupted
rm -rf ./lancedb/
npx gutty nutrition-index && npx gutty gi-index && npx gutty fodmap-index

Common Errors

• "No text content in Gemini response": API quota exceeded or network issue
• "No nutrition match found": Normal for ~15% of ingredients, analysis continues
• "Permission denied": Check service account has Vertex AI User role

Performance

• Processing time: ~25 seconds per image
• Nutrition matching: ~85% ingredient coverage
• Database size: ~50MB total (local LanceDB files)
• Network usage: ~2MB per image (Gemini API only)

Data Sources

• Nutrition: UK COFID database (government nutrition standards)
• Glycemic Index: University of Sydney research database
• FODMAP: Monash University clinical research
• Health guidelines: WHO, ACOG, NICHD peer-reviewed recommendations

All health recommendations are rule-based using evidence-based thresholds. No AI interpretation of results.

System Requirements

• Minimum: Node.js 18+, 2GB RAM, 1GB disk, internet connection
• Recommended: Node.js 20+, 4GB RAM, SSD storage

---

MIT © 2025