README.md

metadata

title: ML Services
emoji: 💍
colorFrom: pink
colorTo: red
sdk: docker
pinned: false

💍 Soul Mate Matchmaking API

AI-powered partner recommendation using Supabase data and scikit-learn.

---

🎯 What It Does

1. Takes a user ID (UID) as input
2. Fetches user profile from Supabase
3. Compares with opposite-gender profiles using ML model
4. Returns top compatible matches (0-100% score)

---

🚀 API Endpoint

GET /recommend/{user_id}

Get AI-matched partners for a user.

Query Parameters:

• top_n (optional, default=10): number of matches to return

Response:

{
  "matches": [
    {
      "user_id": "uuid-here",
      "name": "Fatima Ali",
      "age": 26,
      "city": "Karachi",
      "compatibility_score": 87.5,
      "photo_url": "https://..."
    }
  ]
}

Example:

curl http://127.0.0.1:7860/recommend/44d81ff7-93d2-4805-a973-df9a62a99cf2?top_n=5

---

POST /feedback

Record a like/reject action on a match.

Body:

{
  "user_id": "user-123",
  "target_id": "user-456",
  "action": "like"
}

action: "like" or "reject"

Response:

{ "status": "ok" }

---

GET /health

Check if model is loaded.

{ "status": "running", "model": "loaded" }

---

🔧 Local Development

# Install dependencies
pip install -r requirements.txt

# Train model on Supabase data
python train.py

# Start server
python start.py
# or
uvicorn app:app --reload --port 7860

Interactive docs: http://localhost:7860/docs

---

🧠 Model Features

The model encodes 10 profile attributes:

Category	Columns
Numeric (1)	age
Categorical (6)	religion, marital_status, qualification, country, maslak, region_caste
Text (3)	hobbies, personality_traits, preferred_partner_criteria

Technique:

• LabelEncoder for categoricals
• TF-IDF (max_features=20) for text fields
• MinMaxScaler for age
• Cosine similarity via NearestNeighbors

Total feature dimensions: 1 + 6 + (20×3) = 67 features

---

🚢 Deployment to HuggingFace Spaces

# 1. Train locally first
python train.py

# 2. Clone your HF space
git clone https://huggingface.co/spaces/subhan971/ML-services
cd ML-services

# 3. Copy files
xcopy /s /y "D:\soulmate\soul_mate_app_flutter_backup\recommendation_service\*" ".\"

# 4. Ensure model exists
if not exist "models\recommendation_model.pkl" (
    echo Model missing! Run train.py first.
    exit /b 1
)

# 5. Setup LFS
git lfs track "*.pkl"
git add .gitattributes

# 6. Commit & push
git add .
git commit -m "deploy: matchmaking model"
git push origin main

Space URL: https://subhan971-ml-services.hf.space

---

⚙️ Environment Variables

Variable	Description
SUPABASE_URL	Your Supabase project URL
SUPABASE_SERVICE_KEY	Service role key (for read access)
PORT	Server port (default: 7860)

Set before starting the server:

export SUPABASE_URL=https://xxxx.supabase.co
export SUPABASE_SERVICE_KEY=your-service-role-key

---

📊 Training Data

Model is trained on all active profiles from Supabase profiles table.

To retrain after adding new users:

python train.py

The script:

1. Fetches all active profiles from Supabase
2. Preprocesses features
3. Trains NearestNeighbors model
4. Saves to models/recommendation_model.pkl

---

🔄 Flutter Integration

Flutter calls:

final response = await dio.get(
  'https://subhan971-ml-services.hf.space/recommend/$userId',
  queryParameters: {'top_n': 10}
);

Matches are displayed with:

• Name
• Age
• City
• Compatibility % score
• Photo

---

📁 Project Structure

recommendation_service/
├── app.py              # FastAPI server (2 endpoints)
├── train.py            # Training script (Supabase → model)
├── requirements.txt    # Dependencies
├── Dockerfile          # HF Spaces container
├── start.py            # Launch with env vars
├── models/
│   └── recommendation_model.pkl  # Trained model
├── README.md
├── .gitignore
└── .gitattributes      # LFS tracking

---

✅ Requirements Met

• ✅ Fetch user data from Supabase using UID
• ✅ Preprocess: age, religion, preferred_partner_criteria, personality_traits, hobbies, qualification, marital_status, nationality, sect, region/caste, gender
• ✅ Train ML model (TF-IDF + LabelEncoder + NearestNeighbors)
• ✅ Save as .pkl
• ✅ Backend API: /recommend/{user_id} returns ranked matches
• ✅ Flutter integration: AI Match button → API → display results
• ✅ Only opposite-gender matches
• ✅ Compatibility score 0-100%

---

Questions? Check /docs when server is running.