Trainee Decision Scientist
Bengaluru, Karnataka, India
Built a Cox Proportional Hazard (CoxPH) Model model to predict the probability ofsurvival (retention) of Business Development Managers (BDMs) for following 48months.This pilot was crucial as the company was facing an attrition of 40% and theinsurance industry relies heavily on the relationship between BDMs & customers.Modeled churn based on their performance and incentives, types of banks theywere associated with (urban/rural), proximity of banks from home and otherfeatures, with a back tested accuracy of 70%.Led a team of 12 analysts to design a custom solution for Citi TechnologyInfrastructure (CTI) with the aim to simplify the provisioning, building, andmonitoring of ML sandboxes, utilizing AutoML tools like DriverlessAI and DataRobot.Collaborated closely with VPs and Data Scientists to understand their needs anddesign a self-serve provisioning infrastructure. Once an instance is provisioned,users can leverage AutoML tools to test and build models, and track them using anintegrated MLOps pipeline.This reduced the average time to production for a model from 45 days to 20 days.Also worked alongside DataRobot and DriverlessAI product teams to developtemplates and custom modules tailored to our use case.Inbound Lead Time Prediction: Developed an XGBoost model to predict inboundlead time, reducing stockouts by 30% with 92% accuracy. Implemented forWalmart across 5 countries, results displayed on a dashboard for efficientinventory management.Lost Sales due to On-Shelf Availability: Created a SQL framework to estimate andtrack lost sales from products being unavailable on shelves despite being ininventory, allowing better monitoring of lost sales opportunities
