As team lead for Forecasting Runtime Team, led the design, development, and launch of Metric Evaluation Service (MES) capable of rapidly evaluating ML model prototypes and evaluating them in a standardized way similar to Kaggle. This reduced the time required to evaluate a new ML model from 6 months to 2 weeks. Using this service, paired with the research science team to successfully launch the first deep learning model for Amazon Retail supply chain planning. This launch resulted in the most significant accuracy improvement for item-level demand forecasts in five years (mean 23% accuracy improvement for items fulfilled by Amazon).As team lead for Forecasting Vending Team, led the design and V0 launch of a hierarchical vending system capable of supporting fine-grained regional supply chain planning. This cross-team effort removed demand forecasting as a bottleneck for supporting faster Amazon Retail delivery times (e.g. sub-same day deliveries for Prime members and sub-same hour deliveries for Prime Now).3 years of experience leading team of 7 engineers, and 1 year of experience of cross-team leadership for group comprising of 15 engineers. Led over 150 interview loops as an Amazon Bar Raiser and actively participated in recruiting.Evaluated over 30 promotion proposals as a Calibrated Reviewer.Mentored a total of 10 entry and mid-level engineers.

Led the design, development, and launch of Workflow Orchestration Service and drove its adoption by Amazon's Supply Chain Optimization Team (SCOT) to orchestrate complex ensemble forecasting models. This migration reduced the launch timeline for new physical regions by 6 months (as compared to previous job scheduler-based solution), allowing Amazon Retail to launch in new marketplaces at a rate not previously possible.Completed design and proof-of-concept for distributed computing framework on AWS EMR. This enabled the migration of SCOT's compute workload from self-managed Hadoop clusters to more maintainable AWS-managed EMR clusters, resulting in a net 65% reduction in hardware-related tickets for the SCOT Runtime team and a $100,000 reduction in annual hardware costs for SCOT org.Acted as a mentor for 3 interns and 2 college hires.Conducted interviews and participated in recruiting events.

Paired one-on-one with machine learning scientists to develop ML models capable of forecasting item-level customer demand at Amazon scale.Independently developed an launched an auditor module for the Demand Forecasting team, a fail-safe used to detect and prevent the release of volatile forecast results. This component enabled forecasts to be automatically converted into purchase orders without manual product manager approval, contributing to Amazon Retail's goal of full supply chain automation.Contributed to a linear-regression model used to produce a one-year forecast of customer demand for ~10 million individual Amazon retail items on a daily cadence.Contributed to a random forest model used to predict customer demand for newly launched Amazon retail items with limited historical demand data.

Developed RESTful web services to support IoT networks with applications in environmental monitoring and public safety/emergency services.

Worked as a research assistant in the Robotic Sensor Networks Lab, developing path finding algorithms and artificial intelligence systems.

Research Assistant and programmer for the Konstanz Information Miner (KNIME), a free open-source tool for the development of machine learning, data mining, and data analytics applications with a user experience accessible to non-technical users.Developed module for visualizing the prediction process of complex neural networks.

Developed software infrastructure for personal finance.