Pune EstateLens — AI-Powered Property Price Predictor

🌐 Live Demo: ai-powered-property-price-predictor.vercel.app ↗
⚙️ API: estatelens-backend.onrender.com/docs ↗

---

📋 What is Pune EstateLens?

Pune EstateLens is a production-grade, full-stack AI web application that delivers real-time residential property price predictions across Pune's two fastest-growing real estate corridors. The system combines an XGBoost machine learning model with a modern React frontend, a FastAPI REST backend, and MongoDB Atlas for persistent prediction analytics.

Built as a complete end-to-end ML product — from data generation and model training to cloud deployment with a premium UI.

---

🏘️ Covered Corridors

Corridor	Direction	Key Localities
Dehu Road → Solapur Road	West → Southeast	Dehu Road, Kiwale, Ravet, Hadapsar, Manjri, Loni Kalbhor, Uruli Kanchan
Kolhapur Road → Nashik Road	South → North	Katraj, Khed Shivapur, Bhosari, Chakan, Moshi, Alandi

These two corridors represent Pune's highest infrastructure investment zones (metro, IT parks, ring roads), making them ideal for property price modeling.

---

✨ Features

🏠 1. AI Valuation Engine (Valuation Page)

The core feature. Enter property specifications and receive an instant AI-generated valuation.

Inputs accepted:

• Growth corridor (Dehu-Solapur vs Kolhapur-Nashik)
• Locality (dynamically populated based on corridor)
• BHK configuration (1–4 BHK)
• Total area in sq.ft. (300–3500)
• Number of bathrooms (1–4)
• Floor level (0–30)
• Amenities: Parking, Gym, Swimming Pool, Garden, Security, Clubhouse

Output delivered:

• 💰 Predicted price in ₹ Lakhs (e.g., ₹72.5L)
• 📏 Price per sq.ft. (e.g., ₹7,631/sqft)
• 🎯 Confidence band (±8% range: e.g., ₹66.7L – ₹78.3L)
• 📊 Market comparison charts (BHK trends, locality benchmarks, amenity impact)

🗺️ 2. Interactive Map View (Map Page)

• Leaflet.js powered dark-mode map with CartoDB DarkMatter tiles
• Toggle between both corridors with visual route overlays
• Custom glowing markers for each locality
• Click markers to see locality-level price statistics

🔄 3. Corridor Comparison (Compare Page)

• Configure two properties independently, each with their own corridor and locality
• Side-by-side valuation results with price, per-sqft rate, and confidence bands
• Comparison Dashboard appears after both predictions:
  • Summary banner with price difference (₹L and %), rate comparison
  • Bar chart comparing predicted prices
  • Radar chart comparing feature profiles (BHK, area, floor, amenities)
  • Confidence band overlap visualization
  • Full factor breakdown table with impact explanations
  • "Why the price difference?" dynamic explanation section

📜 4. Prediction History (History Page — MongoDB)

• All predictions are automatically logged to MongoDB Atlas
• Scatter chart showing Price vs. Area correlation colored by corridor
• Side-by-side corridor comparison cards with average predicted price, count, min/max
• Full sortable prediction table with timestamp, locality, BHK, floor, amenities, price

🔁 5. Model Administration (Data & Model Page)

• Upload a new .csv dataset to retrain the XGBoost model live
• Safe Fallback system: staging environment used before live swap — if training fails, previous model stays live (zero downtime)
• Post-retraining metrics (R², RMSE) displayed instantly
• Required CSV schema documented in the UI

🎨 6. Premium Prop-Tech UI/UX

• Immersive 100vh scrolling hero landing page
• Professionally designed Charcoal and Amber dark mode aesthetic
• Seamless glassmorphism interface components with interactive hover states
• Built-in "Market Aesthetics" luxury architectural gallery

---

🧠 Machine Learning System (Deep Dive)

Dataset

The model is trained on a synthetic dataset of 2,000+ Pune residential property records generated using realistic pricing distributions based on Pune's actual real estate market dynamics (corridor premiums, locality tiers, amenity multipliers, floor pricing).

Field	Type	Description
corridor	categorical	dehu_solapur or kolhapur_nashik
locality	categorical	One of 13 localities
bhk	integer	Bedroom count (1–4)
sqft	integer	Total carpet area (300–3500)
bathrooms	integer	Bathroom count (1–4)
floor	integer	Floor level (0–30)
parking	binary	0 or 1
gym	binary	0 or 1
swimming_pool	binary	0 or 1
garden	binary	0 or 1
security	binary	0 or 1
clubhouse	binary	0 or 1
amenities_score	integer	Sum of all amenity flags (derived)
price_lakhs	float	Target variable

Preprocessing Pipeline

Built using scikit-learn's Pipeline + ColumnTransformer:

Raw Input (13 features)
        │
        ▼
┌──────────────────────────────────────┐
│         ColumnTransformer            │
│  ┌──────────────────┐  ┌──────────┐  │
│  │  OneHotEncoder   │  │ Standard │  │
│  │ (corridor,       │  │ Scaler   │  │
│  │  locality)       │  │(numeric  │  │
│  └──────────────────┘  │ features)│  │
│                        └──────────┘  │
└──────────────────────────────────────┘
        │
        ▼
┌──────────────────────────────────────┐
│         XGBRegressor                 │
│  n_estimators=500, max_depth=7       │
│  learning_rate=0.05, subsample=0.8   │
│  colsample_bytree=0.8                │
│  reg_alpha=0.1, reg_lambda=1.0       │
└──────────────────────────────────────┘
        │
        ▼
    prediction (₹ Lakhs)

Model Performance

Metric	Train	Test
R² Score	~0.97	~0.91
RMSE	~5.2L	~8.4L
MAE	~3.8L	~6.1L
5-Fold CV R²	0.91 ± 0.02	—

Why XGBoost?

Property	Benefit
Handles mixed feature types	Both categorical (locality, corridor) and numeric (sqft, floor) natively
Non-linear relationships	Floor-price interactions, locality premiums, amenity stacking effects
Robust to outliers	Real estate data inherently has premium outliers
Feature importance	Can explain which features drive the price most
Fast inference	Sub-millisecond predictions suitable for real-time API responses

Confidence Band

The system reports a ±8% confidence interval around every prediction. This is the empirically observed MAPE (Mean Absolute Percentage Error) of the model on the test set, providing users with realistic uncertainty bounds rather than false precision.

---

🛠️ Tech Stack

Layer	Technology	Version	Purpose
Frontend Framework	React	19	Component-based SPA
Build Tool	Vite	8	Dev server + production bundler
Styling	Tailwind CSS	3	Utility-first design system
Charts	Recharts	3	Interactive data visualization
Maps	React Leaflet	5	Corridor map with markers
Icons	Lucide React	1.8	Consistent icon system
Routing	React Router	7	SPA client-side routing
Backend	FastAPI	0.115	Async REST API
ML Model	XGBoost	2+	Gradient boosted regression
ML Pipeline	Scikit-learn	1.8	Preprocessing + pipeline
Database	MongoDB Atlas	—	Prediction log persistence
DB Driver	PyMongo	4.16	MongoDB Python driver
Data	Pandas + NumPy	—	Data manipulation
Deployment	Render + Vercel	—	Cloud hosting

---

📡 REST API Reference

Base URL (production): https://estatelens-backend.onrender.com

Method	Endpoint	Auth	Description
GET	/api/health	None	Server health + MongoDB status
GET	/api/meta	None	Available corridors and localities
POST	/api/predict	None	Get AI price prediction
POST	/api/compare	None	Compare same property across corridors
GET	/api/stats	None	Market statistics (avg price per locality, BHK)
GET	/api/amenity-impact	None	Price premium per amenity type
GET	/api/history	None	Last 50 predictions from MongoDB
GET	/api/history/corridor-stats	None	Corridor aggregation from history
POST	/api/retrain	None	Upload CSV to retrain model

Sample Predict Request

POST /api/predict
{
  "corridor": "dehu_solapur",
  "locality": "Hadapsar",
  "bhk": 2,
  "sqft": 950,
  "bathrooms": 2,
  "floor": 5,
  "parking": 1,
  "gym": 1,
  "swimming_pool": 0,
  "garden": 0,
  "security": 1,
  "clubhouse": 0
}

Sample Predict Response

{
  "predicted_price_lakhs": 72.5,
  "price_per_sqft": 7631.0,
  "corridor": "dehu_solapur",
  "locality": "Hadapsar",
  "confidence_band": { "low": 66.7, "high": 78.3 },
  "input_summary": {
    "bhk": 2,
    "sqft": 950,
    "bathrooms": 2,
    "floor": 5,
    "amenities_score": 3
  }
}

---

🗂️ Project Structure

AI-Powered-Property-Price-Predictor/
│
├── backend/                          # FastAPI Python backend
│   ├── data/
│   │   ├── generate_dataset.py       # Synthetic data generator
│   │   └── pune_properties.csv       # 2000+ row training dataset
│   ├── model/
│   │   ├── train_model.py            # XGBoost training pipeline
│   │   ├── xgb_pipeline.joblib       # Serialized trained model (2.1 MB)
│   │   ├── model_meta.json           # Corridors & localities metadata
│   │   └── metrics.json              # Training & evaluation metrics
│   ├── database.py                   # MongoDB connection module (graceful degradation)
│   ├── main.py                       # FastAPI app with all endpoints
│   ├── requirements.txt              # All Python dependencies
│   └── .env                          # Local env vars (gitignored — never committed)
│
├── frontend/                         # React + Vite frontend
│   ├── public/
│   │   └── favicon.svg
│   ├── src/
│   │   ├── components/
│   │   │   ├── Header.jsx            # App header with branding
│   │   │   ├── Footer.jsx            # Footer with GitHub CTA
│   │   │   ├── Dashboard.jsx         # Market charts component
│   │   │   ├── PredictionForm.jsx    # Property input form
│   │   │   └── PredictionResult.jsx  # Result display card
│   │   ├── pages/
│   │   │   ├── PredictPage.jsx       # Main valuation page
│   │   │   ├── MapPage.jsx           # Interactive map with corridors
│   │   │   ├── ComparePage.jsx       # Corridor comparison with visualizations
│   │   │   ├── HistoryPage.jsx       # MongoDB prediction history
│   │   │   └── AdminPage.jsx         # Model retraining dashboard
│   │   ├── api.js                    # Centralized API service layer
│   │   ├── App.jsx                   # Root with routing & navigation
│   │   ├── index.css                 # Design system (Tailwind + custom tokens)
│   │   └── main.jsx                  # Entry point
│   ├── vercel.json                   # SPA routing rewrites for Vercel
│   ├── package.json
│   └── vite.config.js
│
├── sample_data.csv                   # CSV template for model retraining
├── .gitignore
└── README.md

---

🚀 Local Development Setup

Prerequisites

• Python 3.9+
• Node.js 18+
• MongoDB Atlas account (free M0 cluster)

1. Clone

git clone https://github.com/ranvirdeshmukh2004/AI-Powered-Property-Price-Predictor.git
cd AI-Powered-Property-Price-Predictor

2. Backend

# Install dependencies
pip install -r backend/requirements.txt

# Configure MongoDB (edit this file with your Atlas URI)
# backend/.env → MONGODB_URI=mongodb+srv://user:pass@cluster.mongodb.net/?appName=Cluster0

# Start API server
cd backend
python3 -m uvicorn main:app --reload --port 8000

Backend runs at: http://localhost:8000 | Swagger UI: http://localhost:8000/docs

3. Frontend

cd frontend
npm install
npm run dev

Frontend runs at: http://localhost:5173 (or next available port)

---

🌐 Cloud Deployment

Backend → Render

Field	Value
Root Directory	backend
Build Command	pip install -r requirements.txt
Start Command	uvicorn main:app --host 0.0.0.0 --port $PORT
Environment Var	MONGODB_URI = your Atlas URI

Frontend → Vercel

Field	Value
Root Directory	frontend
Framework	Vite (auto-detected)
Build Command	npm run build
Output Directory	dist
Environment Var	VITE_API_BASE_URL = https://your-render-url.onrender.com/api

---

🔒 Security & Design Decisions

Decision	Rationale
.env is gitignored	MongoDB credentials never enter version control
MongoDB write is fire-and-forget	A DB failure never blocks or delays the prediction response
Safe fallback retraining	Staging environment used before live swap — guarantees zero downtime
CORS allow_origins=["*"]	Permissive for demo; should be locked to Vercel domain in production
XGBoost serialized with joblib	Committed to repo so Render doesn't need to retrain on every deploy

---

Pune EstateLens — Production-Grade AI Property Intelligence
Live App ↗ • API Docs ↗ • Source Code ↗