Azam-Rabiee/wealthy-estimator-api

Wealth Potential Estimator API

A machine learning service that estimates a user's potential net worth based on a submitted selfie image. The API analyzes facial features and compares them to a database of wealthy individuals to provide wealth potential estimates.

Live URL of the Endpoint

• —Repo (Github): https://github.com/AzamRabiee/wealthy_estimator
• —Repo (HF): https://huggingface.co/spaces/Azam-Rabiee/wealthy-estimator-api/tree/main
• —Swagger: https://azam-rabiee-wealthy-estimator-api.hf.space/docs

🏗️ Architecture

Overview

This project implements a Wealth Potential Estimator using:

• —FastAPI for the REST API framework
• —Hugging Face Transformers for pre-trained vision models
• —PyTorch for deep learning inference
• —Docker for containerization and deployment

Key Components

1. 1.Vision Model: Uses microsoft/resnet-50 for feature extraction
2. 2.Embedding Extraction: Converts selfie images to high-dimensional feature vectors
3. 3.Similarity Computation: Uses cosine similarity to find similar wealthy profiles
4. 4.Wealth Estimation: Weighted average based on similarity scores

Data Flow

Selfie Image → Preprocessing → Feature Extraction → Similarity Search → Wealth Estimation → API Response

🚀 Quick Start

Prerequisites

• —Docker and Docker Compose
• —Python 3.9+ (for local development)

Using Docker (Recommended)

1. 1.Clone and build:

git clone <repository-url>
cd wealth-estimator-api
docker-compose up --build

1. 1.Access the API:
2. 2.API Documentation: http://localhost:8000/docs
3. 3.Health Check: http://localhost:8000/health
4. 4.Predict Endpoint: http://localhost:8000/predict

Local Development

1. 1.Install dependencies:

pip install -r requirements.txt

1. 1.Run the application:

uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

📡 API Endpoints

POST /predict

Predicts wealth potential from a selfie image.

Request:

• —Method: POST
• —Content-Type: multipart/form-data
• —Body: Image file (JPEG, PNG, etc.)

Response:

{
  "estimated_net_worth": 50000000,
  "currency": "USD",
  "similar_profiles": [
    {
      "name": "Elon Musk",
      "net_worth": 250000000000,
      "occupation": "Entrepreneur",
      "similarity_score": 0.85
    }
  ],
  "confidence_score": 0.75
}

GET /health

Health check endpoint.

GET /

Root endpoint with API information.

🔧 Configuration

Key configuration options in app/config.py:

• —MODEL_NAME: Pre-trained model to use (default: "microsoft/resnet-50")
• —MAX_IMAGE_SIZE: Input image size (default: 224)
• —DEVICE: Computation device (default: "cpu")
• —TOP_K_SIMILAR: Number of similar profiles to return (default: 3)

🧠 Model Details

Model Selection

• —Model: microsoft/resnet-50
• —Reasoning:
• —Pre-trained on ImageNet with strong feature extraction capabilities
• —Lightweight and fast for inference
• —Good balance between accuracy and performance

Feature Extraction Process

1. 1.Image Preprocessing: Resize to 224x224, normalize with ImageNet stats
2. 2.Feature Extraction: Use ResNet-50's last hidden state
3. 3.Embedding Normalization: L2 normalization for consistent similarity computation

Similarity Computation

• —Metric: Cosine Similarity
• —Reasoning:
• —Scale-invariant similarity measure
• —Works well with normalized embeddings
• —Computationally efficient

Wealth Estimation Algorithm

estimated_net_worth = Σ(similarity_score × profile_net_worth) / Σ(similarity_scores)

📊 Dataset

The wealthy profiles dataset (data/wealthy_profiles.json) contains:

• —10 mock wealthy individuals
• —Each profile includes: name, net worth, occupation, and pre-computed embeddings
• —Net worth values range from $65B to $250B

Note: This is a demonstration dataset. In production, you would:

• —Use real wealthy individual data
• —Implement proper data privacy measures
• —Use more sophisticated embedding generation

🏗️ Engineering Decisions

Framework Choices

• —FastAPI: Modern, fast, automatic API documentation
• —PyTorch: Industry standard for deep learning
• —Transformers: Easy access to pre-trained models

Performance Optimizations

• —Model loading at startup (not per request)
• —Image preprocessing optimization
• —Efficient similarity computation

Error Handling

• —Comprehensive input validation
• —Graceful error responses
• —Detailed logging

Security Considerations

• —Input file validation
• —Non-root Docker user
• —Health checks for monitoring

🧪 Testing

Manual Testing

1. 1.Use the interactive API docs at /docs
2. 2.Upload a selfie image
3. 3.Verify the response format and values

Example cURL Request

curl -X POST "http://localhost:8000/predict" \
     -H "accept: application/json" \
     -H "Content-Type: multipart/form-data" \
     -F "file=@your_selfie.jpg"

🚀 Deployment

Docker Deployment

# Build and run
docker-compose up --build

# Run in background
docker-compose up -d

Cloud Deployment Options

• —Heroku: Use the provided Dockerfile
• —AWS ECS: Deploy using docker-compose
• —Google Cloud Run: Container-based deployment
• —Azure Container Instances: Serverless container deployment

📝 Assumptions

1. 1.Model Accuracy: This is a demonstration project - accuracy is secondary to robustness
2. 2.Wealth Correlation: Assumes facial features correlate with wealth potential (for demo purposes)
3. 3.Dataset: Uses mock wealthy profiles - not real data
4. 4.Privacy: No actual personal data is stored or processed beyond the request
5. 5.Scalability: Designed for moderate traffic - can be scaled with load balancers

🔮 Future Improvements

1. 1.Model Enhancement:
2. 2.Fine-tune on wealth-related datasets
3. 3.Use ensemble models for better accuracy

1. 1.Data Pipeline:
2. 2.Real-time wealthy profile updates
3. 3.More diverse dataset

1. 1.API Features:
2. 2.Batch processing
3. 3.Authentication and rate limiting
4. 4.Caching for similar requests

1. 1.Monitoring:
2. 2.Request/response logging
3. 3.Performance metrics
4. 4.Error tracking

📄 License

This project is for demonstration purposes only.

🤝 Contributing

This is a demonstration project for a coding interview. For production use, please implement proper security, privacy, and ethical considerations.