Manage team within Measurement Ad Tech Data Science (MADS) central science.

Manage the Specialized Selection team as a part of the Supply Chain Organization Technology (SCOT) organization. We are a highly cross-functional team of ~20 across science, engineering, product, and analytics. We build the models and system that determine which products Amazon should offer to customers in our fastest delivery formats, using a variety of scientific tools including: machine learning (both supervised and unsupervised), nonconvex optimization, natural language processing, reinforcement learning, and causal inference.

Collaborated with scientists and engineers to improve rainfall estimation algorithms and implement in production, receiving a patent for this work.Also developed machine learning models to leverage weather time series data to improve yield predictions;, collaborating with the product team to design mechanism to communicate information to growers.

Research focused on applying machine learning algorithms to improve forecasts of rare, high-impact weather events including extreme rainfall; severe weather, including tornadoes, hail, and high winds; and reduced cloud ceiling and visibility. Primary project involved building a new automated excessive rainfall outlook (ERO) based on a probabilistic machine learning model for use by NOAA Weather Prediction Center (WPC) forecasters. The ERO product is used by emergency managers and others to understand risk of flash flooding in their area over the next 1-3 days, based on daily risk outlooks. Worked with forecasters to understand existing pain points with the forecasting process, design a model that would cater to their needs, evaluated the model both objectively and subjectively in a testbed setting, and iterated on the model based on customer feedback to further improve it. The model rolled out operationally in 2019 and is currently used operationally by WPC forecasters to generate their ERO.Other research areas included (1) interpretable ML and the dynamical insights implied by the ML models; (2) applying extreme value statistics to quantify flash flood observational characteristics, including comparing the extremes of several prominent observational datasets; (3) forecast verification techniques for extremes and spatiotemporal datasets; and (4) interdisciplinary research with on forecast perception and decision-making in multi-hazard scenarios. Other highlights:• Published six first author and fifteen total journal articles• Authored three grant proposals, all funded; wrote statement of work for two of them. Total award ~$700K• Presented over 20 conference abstracts and/or invited talks• Short course co-instructor/organizer, Machine Learning in Python for Environmental Science Problems• Participant in six field programs; led forecast operations planning and execution in one• Reviewer of over ten manuscripts from at least six different journals