CredX | ML-Driven Fair Credit Scoring

CredX is a next-generation credit evaluation platform designed to bridge the gap for the "credit-invisible." By utilizing a 5-Layer Fairness Architecture and alternative behavioral data, it eliminates systemic bias while providing 85% predictive accuracy for users without traditional credit histories.

---

Project Showcase

Live demonstration of the FairScore™ engine and bias mitigation dashboard.

---

Key Features

• 5-Layer Fairness Architecture: Systematic mitigation of representational, measurement, algorithmic, aggregation, and historical bias.
• Alternative Data Scoring: Leverages UPI, utility patterns, and digital footprints to validate income with 85% accuracy.
• Delphi Consensus Engine: A specialized ensemble model balancing fairness (50%), accuracy (30%), and prediction diversity (20%).
• Explainable AI (XAI): Integrated SHAP explanations for every score to ensure regulatory compliance and user trust.
• Real-Time Bias Detection: Production-grade monitoring that identifies and flags geographic or demographic discrimination instantly.

---

Technical Stack

Frontend

• React.js: Interactive dashboard for data visualization and credit application.
• Tailwind CSS: Modern, crisp UI designed for financial transparency.
• Recharts: High-performance rendering of credit trends and fairness metrics.

Backend & ML

• FastAPI: High-performance Python framework for real-time scoring inference.
• Scikit-Learn & Fairlearn: Core libraries for model training and bias mitigation.
• SHAP: Implementation of Explainable AI for model transparency.
• PostgreSQL/MongoDB: Robust handling of traditional and alternative data streams.

---

Interface Preview

Homepage	Application Form	Credit Dashboard

---

Engineering Deep Dive

1. The Fairness-Accuracy Harmony

A major challenge in fintech ML is the trade-off between model accuracy and fairness. I implemented the Delphi Consensus Engine, which uses multi-objective optimization. This resulted in a 40% improvement in fairness metrics with zero net loss in predictive accuracy.

2. Regional Bias Elimination

Traditional models often penalize users based on their PIN code. CredX uses region-aware feature engineering and adversarial debiasing to ensure that a user's location does not unfairly influence their creditworthiness, promoting geographic financial equity.

---

Getting Started

Prerequisites

• Python 3.9+
• Node.js & npm

1. Backend & ML Setup

cd ML_Model
pip install -r requirements.txt
# Run the FastAPI server
source ./.venv/bin/activate
python main.py

2. Frontend Setup

cd Frontend
npm install
npm run dev