blanc.Data
Projects/Financial Services / Credit Risk

Buy Now, Pay Later (BNPL) Credit Risk Scoring & Loan Recommendation Pipeline

Prototype In Progress

An end-to-end automated credit decisioning engine built to evaluate applicant default risk in real time and dynamically recommend tailored credit limits and loan terms.

Technologies Used:DatabricksPySparkDelta Lake (Medallion Architecture)PySpark MLDatabricks Lakeview DashboardsStreamlit
Core Analytical Question

“Automate credit underwriting for short-term installment loans by predicting default probabilities, balancing approval rates against financial exposure, and eliminating manual review bottlenecks.”

Business Problem & Objectives

Automate credit underwriting for short-term installment loans by predicting default probabilities, balancing approval rates against financial exposure, and eliminating manual review bottlenecks.

Ingested Data Inputs

Kaggle Dataset covering applicant demographic records, raw transaction logs, and historical repayment histories.

Source: Kaggle Dataset

Data Pipeline Architecture

Medallion Structure: Bronze Layer (Raw Ingestion with schema enforcement) → Silver Layer (Cleaning, deduplication & standardization) → Gold Layer (Feature Store: DTI ratios, repayment velocity, delinquency counts, credit utilization) → ML Scoring & Decision Engine

Machine Learning & Statistical Method

Default Probability Scoring: Uses a supervised classification pipeline built in PySpark ML to score each applicant's probability of default (PD). Rules-Based Recommendation Layer: Pairs the ML risk score with business logic to dynamically assign risk tiers, recommend maximum safe credit limits, and output instant Approve / Decline / Manual Review decisions.

Decision Engine & User Interface

Designed a multi-level Databricks Lakeview Dashboard (and an interactive Streamlit prototype) to monitor portfolio-level risk exposure, evaluate cost-loss matrices across probability thresholds, and inspect individual applicant scoring breakdowns.

Analytical Scope: Independent prototype built on public data while preparing to take on client work.

Independent prototype built on public data while we prepare to take on client work. Not produced for or endorsed by any organisation.