PropIntel AI

Collateral Intelligence Engine for NBFC Lending

An AI-powered system that transforms property collateral assessment from a 10-day manual process into a 2-second intelligent decision. Built for Indian NBFCs and lending institutions.

---

Languages & Frameworks & Tools

---

## Overview

The Problem

When an NBFC issues a property-backed loan, determining the true market value of collateral is slow, manual, and expensive. Lenders rely on physical valuators who take 7-10 days, use outdated government circle rates, and provide no insight into how quickly the property could be liquidated in distress. Fraudulent applications with size manipulation or fake documents further expose lenders to risk.

The Solution

PropIntel AI ingests basic property data (address, size, age, configuration) and instantly computes:

• Accurate market and distress valuations
• Liquidity prediction (time-to-sell and resale index)
• Fraud and anomaly detection
• Lender confidence scoring
• Approve / Review / Reject decision output

What took 10 days and manual effort now takes under 2 seconds.

---

Key Features

Feature	Description
AI Property Valuation	Gradient Boosting model predicts market and distress values using circle rates, proximity, and property attributes
Liquidity Intelligence	10-factor engine computes resale index (0-100), time-to-sell, and supply pressure metrics
Fraud Detection	Rule-based anomaly detection flags size manipulation, floor mismatches, price-circle rate divergence, and NPA zone conflicts
Proximity Intelligence	Real geocoding (Nominatim + Overpass API) computes distances to metro, hospitals, schools, IT parks, and highways
Decision Engine	Combines valuation, liquidity, confidence, and fraud signals into a clear Approve / Review / Reject recommendation
Explainable Outputs	Every prediction includes key value drivers, confidence breakdown, and risk flags
PDF Report Generation	Two-page print-ready lender assessment report with all metrics and disclaimers

---

---

Installation and Setup

# Clone the repository
git clone https://github.com/7Vedansh/PropIntel.git
cd PropIntel

# Create and activate virtual environment
python -m venv venv
source venv/bin/activate        # Linux/Mac
# venv\Scripts\activate         # Windows

# Install dependencies
pip install -r requirements.txt

# Generate synthetic training data
python data/generate_data.py

# Train the valuation model
python train.py

# Start the FastAPI backend
uvicorn api.main:app --reload --port 8000

# In a new terminal, start the Streamlit dashboard
streamlit run app/main.py

The API documentation is available at http://localhost:8000/docs once the backend is running.

---

System Architecture

---

Workflow

Step 1: User enters property address + basic details
            |
Step 2: Address geocoded to lat/long (5-strategy fallback)
            |
Step 3: Proximity distances computed (Overpass -> Nominatim -> Fallback)
            |
Step 4: Circle rate looked up from government database
            |
Step 5: Feature engineering assembles all inputs
            |
Step 6: Valuation model predicts market + distress value
            |
Step 7: Liquidity engine computes resale index + time-to-sell
            |
Step 8: Fraud engine runs 9 anomaly detection rules
            |
Step 9: Confidence engine scores data quality + signal agreement
            |
Step 10: Decision engine generates Approve/Review/Reject + safe loan amount

---

Project Structure

PropIntel/
├── api/                          # FastAPI backend
│   ├── __init__.py
│   └── main.py                   # REST API endpoints + decision logic
│
├── app/                          # Streamlit frontend
│   ├── main.py                   # Premium dashboard UI
│   └── components/
│       └── python_report.py      # PDF report generator
│
├── engine/                       # Core intelligence engines
│   ├── __init__.py
│   ├── valuation.py              # ML valuation (Gradient Boosting)
│   ├── liquidity_v2.py          # 10-factor liquidity scoring
│   ├── fraud.py                  # 9-rule anomaly detection
│   ├── confidence.py             # 3-signal confidence scoring
│   ├── geocoder.py               # 5-strategy address geocoding
│   └── proximity.py              # Overpass + Nominatim distance engine
│
├── data/                         # Data layer
│   ├── circle_rate_db.py         # Government circle rate database
│   ├── generate_data.py          # Synthetic data generator (500 rows)
│   └── synthetic_properties.csv  # Generated training dataset
│
├── models/                       # Trained model artifacts
│   ├── __init__.py
│   ├── valuation_model.pkl       # Serialized Gradient Boosting model
│   ├── feature_names.pkl         # Model feature list
│   └── model_metadata.pkl        # Training metadata
│
├── notebooks/
│   └── model_training.ipynb      # Jupyter training + validation notebook
│
├── train.py                      # Model training pipeline
├── requirements.txt               # Python dependencies
├── Dockerfile                     # Container configuration
└── README.md

---

Business Impact

Metric	Before	After
Assessment time	7-10 days	< 2 seconds
Cost per assessment	~10,000 INR	Near zero (marginal compute)
Fraud detection	Manual spot-checks	Automated 9-rule engine
Liquidity insight	None	Resale index + time-to-sell
Decision consistency	Varies by valuator	Standardized AI output
Scalability	Limited by manpower	Unlimited concurrent assessments

Input Property Details

## Report Generated ## Valuation Analysis ## Liquidity Analysis ## Proximity Analysis ## Market Intelligence Analysis

Author

Vedansh & Ameya - GitHub

Team Arjuna