ML CREDIT RISK • PRODUCTION PIPELINE

Understand your loan eligibility before you apply.

Finsta QuickLoans connects the application directly to the calibrated Probability of Default model trained on the full 100,000-row dataset.

Indicative Interest Rate

12.00% – 36.00% p.a.

Risk-based indicative pricing. Final lending terms are subject to underwriting.

My Project Journey

I built Finsta QuickLoans as my personal fintech project to bring together finance, credit-risk analysis and machine learning in a simple quick-loan experience. The project uses a historical loan dataset to train a credit-risk model, then combines the model's Probability of Default with an affordability and pricing policy to produce an indicative offer.

How Finsta QuickLoans works

1. Enter your application details01

2. Calculate affordability baseline02

3. ML model estimates default risk03

4. Generate indicative offer04

Browser ML model

Check Loan Eligibility

These are the application-time details used by the Finsta QuickLoans browser-compatible credit model.

Applicant Information

These inputs map directly to the raw fields expected by the browser ML model. Derived features are rebuilt in the browser using the same credit_features.py formulas used during training.

Age

Gender (audit field; not a model feature)

Marital Status

Dependants

Education Level

City Tier

Residence Type

Current Job Industry

Employment Type

Current Job YOE

Total YOE

Last Month Salary (₹)

Other Monthly Income (₹)

Credit Score

Net Worth (₹)

Number of Credit Cards

Credit Card Utilisation (%)

Existing Loans

Existing Monthly EMI (₹)

Requested Loan Amount (₹)

Tenure (days)

Model integrity: Gender is retained for auditability but excluded from prediction. Interest rate is also excluded because it is a pricing output, not an application-time model input.

Production ML Probability of Default

0.00%

Loading model metadata…

—

Decision—

PD-Based Rate—

Offered Rate—

Approval Policy—

Loan Requested—

Tenure—

—

Why this matches the notebook

Browser → model.json/model.bin → PD → notebook pricing/decision logic

The production model is the Stage-2 model retrained on the full dataset, not the Stage-1 80/20 test model. Feature engineering, categorical encoding, tree traversal and isotonic calibration are reproduced in the browser from model.json + model.bin.

My Loans

Your accepted applications and completed loans will appear here.

No active loan

Accept an eligible offer to create an application.

READY

Completed Loan — Demo

₹30,000 • 6 months • Fully repaid

COMPLETED