GoldFi: An Intelligent Risk Control Platform for Gold Loans Combining Machine Learning and Rule Engines

GoldFi is an intelligent risk control platform designed specifically for gold loan scenarios. It combines a 70% machine learning prediction component with a 30% rule engine to achieve explainable, real-time, and accurate credit decisions. This hybrid approach balances predictive flexibility and business certainty, addressing pain points in traditional manual risk assessment for gold loans in emerging markets.

In emerging markets like India, gold loans are a crucial part of inclusive finance, supporting millions of small businesses and individuals. However, traditional risk control relies on manual review, which is inefficient and hard to scale. GoldFi aims to solve this by providing an AI-driven credit risk prediction platform for banks and NBFCs, enabling real-time, intelligent lending decisions.

GoldFi's key innovation is its Hybrid Risk Engine: Final Risk Score = (ML Probability ×0.7) + (Logical Risk Score ×0.3)

Model Selection

Logistic Regression (ROC-AUC:95.18%) was chosen over XGBoost (95.09%) due to better explainability and easier deployment, despite minimal performance difference.

Feature Engineering

Input includes 12 core dimensions: borrower occupation code, historical default records, active loans count, total debt burden, interest rate, loan value ratio (LTV), monthly payment, CIBIL credit score, monthly income, work experience, income-to-payment ratio, loan amount and term.

First Layer: Rule Engine Scoring

Penalty points for:

• CIBIL <600 (+25 risk points)
• Debt-to-income ratio >50% (+20 risk points)
• LTV >85% (+15 risk points)

Second Layer: Hard Constraints

Any of these triggers high risk (score ≥85):

• Monthly payment strictly greater than monthly income
• Loan amount higher than gold valuation
• CIBIL score below 600

Third Layer: Explainable AI (XAI)

Generates human-readable insights (e.g., "high LTV ratio", "income deficit", "poor credit score") to meet regulatory transparency requirements and assist manual review.

Frontend

• HTML5/CSS3: Glassmorphism design language
• ES6 JavaScript: DOM operations, form validation, dynamic dashboards, async API calls
• Micro-interactions: Pulse loading, smooth SVG progress dials, hover animations

Backend

• FastAPI: High-performance async framework
• Uvicorn: ASGI server
• Pydantic: Data validation & configuration
• Scikit-Learn: Model training/inference
• Joblib: Model serialization

Deployment

• Frontend: GitHub Pages
• Backend: Render cloud platform
• API Docs: Auto-generated Swagger/OpenAPI docs

GoldFi integrates a real-time gold valuation feature. Users input gold weight and purity to get instant collateral value estimates, digitizing physical asset assessment and shortening loan approval time.

GoldFi offers key takeaways for AI financial applications:

1. Hybrid architecture is superior to pure model solutions: Balances AI prediction and business logic certainty in regulated finance.
2. Explainability is a prerequisite for production deployment: Critical for trust and compliance in credit decisions.
3. Engineering tradeoffs matter more than model complexity: Choosing Logistic Regression over XGBoost prioritizes maintainability and speed.
4. Scenario-specific design enhances value: Gold valuation and LTV monitoring are tailored to gold loan needs.

GoldFi represents a pragmatic AI financial application. It prioritizes reliability, transparency, and ease of deployment over complex models. Its hybrid approach (combining ML and traditional risk control wisdom) is well-suited for inclusive finance digitalization and likely to gain trust from financial institutions.