Data Scientist

Designed and implemented a data-driven homepage tile optimization system, dynamically adjusting placements based on user engagement patterns;

Aggregated and analyzed time-series data on impressions, clicks, and "Add to Bag" actions to uncover user behavior trends;

Engineered advanced features (e.g., temporal trends, engagement metrics) to enhance the predictive accuracy of optimization models;

Developed and validated an XGBoost model to predict "Add to Bag" rates, using Mean Absolute Error (MAE) and R-squared for performance assessment;

Integrated Thompson Sampling (Multi-Armed Bandit) for real-time tile placement optimization, adapting dynamically to user interactions;

Automated the optimization process, ensuring continuous improvements without manual intervention.

Monitored and fine-tuned the model’s performance, achieving a 15% increase in "Add to Bag" actions and directly boosting revenue;

Led a team of data scientists to develop an unsupervised learning framework for visitor segmentation using Adobe Clickstream data;

Designed and operationalized a bi-weekly MLOps pipeline to automate data ingestion, feature engineering, model retraining, and performance monitoring;

Enhanced targeted marketing campaigns by implementing data-driven visitor segmentation, improving personalization and engagement;

Optimized marketing budget allocation through intelligent resource distribution, maximizing ROI;

Integrated segmentation insights into a conversational Tableau (ThoughtSpot) dashboard for real-time data accessibility, model governance, and strategic decision-making;

Improved conversion rates, A/B testing effectiveness, and marketing efficiency by enabling data-driven decision-making.

Led cross-functional collaboration with model development, deployment, monitoring, risk, and fair lending teams to ensure seamless ML model lifecycle management;

Managed a portfolio of 37 ML models in production, overseeing performance monitoring, recalibration, and compliance adherence;

Developed and automated MLOps pipelines, significantly reducing turnaround time (TAT) for model updates and deployments;

Conducted root cause analysis and model recalibration, improving performance transparency and regulatory compliance;

Implemented workforce optimization strategies, minimizing redundancies and improving operational efficiency;

Enhanced regulatory compliance and fair lending governance, ensuring model transparency and justifiable decision-making;

Designed data-driven storytelling and documentation frameworks, providing clear insights into model decisions for stakeholders.

Developed and deployed an Auto Fraud Detection System using NLP techniques and pre-trained models (BERT, GloVe) to analyze First Notice of Loss (FNOL) data and detect fraudulent claims;

Designed a multiclass claims triaging model to forecast claim urgency and litigation potential, optimizing resource allocation and response times;

Implemented fraud classification models with high accuracy, reducing financial losses and manual investigation efforts;

Predicted claim severity ($) to enable proactive fraud detection and investigation prioritization;

Integrated the system into the claims workflow, enhancing operational efficiency and reducing litigation costs;

Achieved quarterly savings of $250K by improving fraud detection and claims processing efficiency.

Developed classification models to predict customer responses to marketing campaigns and cross-selling initiatives, leveraging behavioral data, demographics, and engagement history;

Applied advanced ML techniques to refine audience segmentation, improving conversion rates and customer engagement;

Implemented MLOps pipelines to automate model development, deployment, monitoring, and recalibration for continuous optimization;

Designed real-time monitoring reports, providing insights into campaign performance, model robustness, and profitability (PnL tracking);

Optimized budget allocation strategies, ensuring marketing spend was maximized for ROI;

Established a structured monitoring system to track long-term model performance and campaign effectiveness.