I am a data scientist and applied econometrician who lives in Seattle, WA. I work as a data science technical lead and manager at Google, and I am a part of the Operations and Infrastructure Data Science (ODS) organization. We are responsible for the forecasting and capacity optimization decisions for Google’s data centers around the world.
Before Google I worked at Microsoft as a data science manager in the Commercial Software Engineering group (CSE). CSE is a global engineering organization that works directly with customers looking to leverage the latest technologies to address their toughest problems and transform their industries using cloud-based solutions. Before CSE I worked as a data scientist in the Office Experience group in Redmond, WA, where I have spent time developing forecasting applications and providing general data science consulting. And before that, I worked in Azure developing forecasting models and tools on top of Azure Machine Learning. This work culminated in the forecasting capabilities of the Azure AutoML service.
And before Microsoft, I was a senior economist at Amazon.com. At Amazon I was one of the founding members of the Economics team, reporting directly to Dr. Patrick Bajari, Amazon’s Chief Economist and Vice President. The mandate of the team was to perform internal consulting services by supplying data-informed answers to key questions from Retail, Marketplace, Prime, Kindle, and AWS businesses. The exact nature of projects was confidential, but methodologically I engaged in econometrics of program evaluation, elastiticity estimation, pricing, and market design. On top of that, I took on multiple forecasting and predictive modeling projects that typically had a stronger data mining focus.
My biggest contribution to Amazon was the work on the Buybox algorithm, which defines the competition rules for Amazon marketplace. It was a perfect illustration of an application where mainstream machine learning approaches were not successful. Such algorithms generally perform best when “ground truth” training data are available. In contrast, when “ground truth” is fundamentally unobservable, it can still be possible to get a lot of mileage on a problem by adopting a principled modeling framework. This is exactly the kind of problem econometrics was developed to handle, so we brought appropriate tools to bear and applied them at scale. I hired and developed a team of data scientists and economists, and collaborated closely with a sister engineering team to develop the model and put it into production. Along the way I had to construct and implement a custom-tailored A/B experimentation framework to enable rapid iteration over candidate models.
Before Amazon I earned a B.S. in economics in 2004 from The Higher School of Economics in Moscow, Russia as well as a M.A. in economics in 2006 from the New Economic School, also in Moscow. Subsequently, I earned a M.A. in 2008 and a Ph.D. in 2011, both in economics and both from the University of Minnesota. My primary research interests are in applied econometrics, industrial organization applied to internet platforms, and machine learning. Details of my professional activity can be found on my Linkedin page.
In my spare time I enjoy hiking around the gorgeous Seattle area, all kinds of water-related activities such as sailing, kayaking, and scuba diving, traveling to U.S. National Parks all over the country, and spending time in the company of dear friends.