
Data Scientist
Mu Sigma · Aug 2022 – Present
- Developed and tested hypothesis for feature selection
- Built and optimized an XGBoost model in Python, improving renewal prediction accuracy across multiple membership segments
- Applied SHAP (SHapley Additive exPlanations) to generate waterfall charts identifying key renewal drivers, translating model outputs into actionable business insights
- Automated the end-to-end pipeline (data ingestion to model scoring) and deployed results via a PowerBI dashboard
- Drove a 3% increase in overall renewal rates from FY24 to FY25, contributing an estimated $400M in retained membership revenue
- Redesigned the lane-planning process to support continuous, year-round operations by leveraging both short-term and long-range demand forecasts
- Broke down high-level demand forecasts into lane-level forecasts using historical data and filling a major gap in the planning process
- Addressed a forecasting gap by incorporating carrier holiday calendars, preventing planning efforts from being allocated to non-operational carrier days
- Added new data points like carrier performance and delivery costs to help the team pick better lanes
- Worked with forecasting and carrier teams to roll out the new process, cutting carrier penalties by 5% and operational costs by 3%
- Built a framework using LLMs to automatically label 200+ company data tables by sensitivity level (Restricted, Confidential, Internal, Public), replacing manual policy review
- Designed a process that masks sensitive data (like personal info and payment details) before sending table samples to the LLM
- Created a log that shows the reasoning and confidence score behind each table's classification, so the governance team can review and trust the results instead of doing it manually
- Built and tested the framework using Claude, achieving strong policy reasoning accuracy for enterprise-scale classification

