Nitin Kumar

Scorecard for Two wheeler Loans

• Generated PD score based on risk profile of customers and pricing eligibility criteria, for model-based customer acquisition and implemented Model based business recommendation to block/increase business volume at dealer level segments for two-wheeler loans. Created settlement framework for unsecured loan portfolio and successfully conducted A-B testing for the same.

Aggregation report generation

• Developed a reliable cron job system for data cleanup and report generation using MySQL and the Moustache rendering engine. Also, set up an email alert mechanism to proactively detect and address spike in errors. TOOLS: CSS, React.js, JQuery.

Bureau based Underwriting Model

• Developed a Random Forest Model (AUC -77%,KS -45%, GINI - 54%)to assess the credit risk of business loan applicants using Python (libraries like Scikit-learn, Pandas).
• Developed RESTful APIs using Django, and FastAPI to serve machine learning models and orchestrated deployment pipelines using Docker and Kubernetes for containerization. Integrated CI/CD pipelines using Jenkins for automated builds, tests, and deployments.

Collection Models

• Developed Bounce Model for pre-EMI calling of non-delinquent customers and reduced overall bounce rate by 1.2% after deployment. Developed Risk of Flow/Rollback Model using Random Forest for 1-90 DPD customers used by collection team in Field allocation and reduced Gross NPA by 0.2%.Developed Logistic Regression based Recovery Model for existing NPA customers and improved monthly recovery rate (amount paid/outstanding) of 180+ DPD customers by 0.4%.

Payment Integration

• Led the implementation of robust reporting mechanisms for EMI and repayments to reduce manual efforts by ~50%, enhancing data visibility compliance with RBI regulations.
• Directed the seamless integration with UPI and Razorpay payment gateway, meticulously ensuring an exceptional High Payment Success Rate, reaching 99.70% . Ensured successful testing and smooth operation of LendingKart MSME loan app after deployment by continuous monitoring application performance, payment success rate and transaction volume. TOOLS: Python, Django, MySQL database, Razorpay SDK, webhooks, GraphQL

• Strengthen Supply of medical essentials using data driven decision making: Utilize data analytics to identify the most affected rural areas using data on COVID-19 case trends, healthcare infrastructure availability and to prioritized resource allocation for effective interventions. Established partnerships with government agencies, private sector, and other NGOs to boost the supply chain for emergency medical essentials.

CCAR Model Validation

• Worked on loan level (Discrete time hazard) PD models in SAS EG using Expected loss (EL) framework to estimate loss forecast for stress scenarios. Performed sensitivity analysis of idiosyncratic attributes as well as macro attributes, model benchmarking to improve model performance and stress testing (9 quarter forecast from Jump off) used by bank to estimate.
• Conducted pre-implementation review to validate conceptual soundness of model and performance evaluation based on modeling data, Design and Performance in adhering to SR 11-7&SR12-17 guidelines released by Federal Reserve.