At Kotak 811, I built a LightGBM model that scores 15 million active customers daily for elevated fraud risk. I used behavioural signals such as PIN changes and balance enquiries to identify pre-fraud patterns.
I developed Clari5/Falcon rules that use vulnerability scores to decline suspicious UPI transactions in real time. The combined model-and-rule framework improved fraud recall by 13 percentage points, with 9% precision on incremental alerts.
At Myntra, I developed inventory forecasts and anomaly detection for product discounts, and built a stock-transfer model that reduced new purchase orders by approximately 11%. Earlier, at Digii formerly CollPoll, I automated analytics workflows with Python and built dashboards for the product.

