🇦🇪 DubaiPropAI

DubaiPropAI is an end-to-end, production-grade Artificial Intelligence application for predicting real estate investment metrics specifically designed for the Dubai property market.

It simulates a full-stack prop-tech architecture utilizing an XGBoost Machine Learning Pipeline, a FastAPI backend, and a dynamic React + TailwindCSS dashboard, completely orchestrated via Docker.

This repository is structured as a monorepo containing multiple decoupled layers:

• Frontend (/fe): React.js, Vite, TypeScript, and TailwindCSS 3. Builds a glassmorphism-styled dashboard serving dynamic metric cards.
• Backend (/be): Python, FastAPI, SQLAlchemy, and Pydantic. Securely exposes REST endpoints for predictions and database mutations.
• Machine Learning (/ml): Python, Pandas, Scikit-learn, XGBoost. Sandbox for generating simulated market data, training prediction models, and generating joblib artifacts.
• Database: PostgreSQL integrated with Alembic for clean structural migrations.
• Containerization: Unified with docker-compose to manage the bridge network linking the React web server, FastAPI inference server, and PostgreSQL database.

The easiest way to bootstrap the entire environment locally is via Docker.

• Docker Desktop (Running)
• Git

1. Clone the repository:

   git clone https://github.com/your-username/DubaiPropAI.git
   cd DubaiPropAI

2. Build and spin up the Docker Cluster:

   docker-compose up --build -d

   This command spins up a Postgres Database (db), builds the FastAPI backend (backend), and launches the NGINX-served React frontend (frontend).

3. Monitor the logs (Optional):

   docker-compose logs -f

Once your Docker containers are actively running, you can interact with the various endpoints natively through your web browser.

• URL: http://localhost
• Description: Play around with the React frontend! Alter the square footage, bedrooms, or Dubai sub-areas like Marina or JVC and get immediate AI-based predictions.

• URL: http://localhost:8000/health
• Description: Quick health check ensuring the backend handles requests and the model is loaded.

• URL: http://localhost:8000/docs
• Description: Test out backend API endpoints (POST /predict, GET /properties) safely without opening the frontend. Provided natively by FastAPI.

The models predict three core investment variables based on inputs (area, property_type, size_sqft, bedrooms):

1. Estimated Price (AED)
2. Current Rental Yield (%)
3. 5-Year ROI (%)

The training pipeline requires a CSV. Out of the box, we use a simulation script to seed robust fake transaction data across the city.

• ml/generate_data.py: Mocks 2,000 realistic listings and exports to ml/data/dubai_properties.csv.
• ml/train_model.py: Ingests the CSV, trains the pipelines, scores the mean absolute errors (MAE), and securely saves to ml/models/models.joblib.

(Both run cleanly out of box, but can be altered whenever you collect scraped real property data!)