Dec 2020 – Oct 2025 · 4 yrs 10 mos

Jul 2020 – Present · 5 yrs 9 mos

• http://workonclimate.org/ - a non-profit focused on mobilizing humanity's talent to build a regenerative green economy, home to the world's largest climate work community.
• We build networks that support and bring together people and organizations leading the regenerative transformation in their field.
• We are currently most known for our community, which is helping 35,000+ people grow into such leaders - turning their work, their company, and their field into a force for regeneration.

Apr 2020 – Jul 2020 · 3 mos

• Volunteering as a software engineer (embedded stuff + QT GUI, all C++) and PM to help build low-cost ventilators for developing countries in the context of COVID-19. We came 3rd place among 200+ teams in the CoVent Challenge.

Jan 2017 – Jan 2019 · 2 yrs

• Helping women enter software engineering.

Feb 2016 – Jun 2016 · 4 mos · Mountain View

• Helping teach coding to a class of middle-school girls at one of Mountain View's schools.

Jan 2013 – Jul 2020 · 7 yrs 6 mos · Mountain View

• Latest: a new programming model for machine learning, with the goal to make ML a programming primitive and make self-tuning of software systems ubiquitous: https://arxiv.org/abs/1810.00619 , https://www.youtube.com/watch?v=Zd5NvRu0Gu0&ab_channel=EugeneKirpichov
• Previous: Jack of all trades at what used to be the MapReduce team - nowadays Flume (internally) and Cloud Dataflow (externally) / Apache Beam (open source)
• dynamic work rebalancing ("liquid sharding" - https://cloud.google.com/blog/big-data/2016/05/no-shard-left-behind-dynamic-work-rebalancing-in-google-cloud-dataflow) and autoscaling - the impact is solving the problem of MapReduce stragglers, and freeing users of the necessity to specify sharding or the amount of resources to allocate to the job.
• IO connectors - enabling new use cases and much higher performance by reframing IO (reading/writing data) as just another data processing task, instead of a programming primitive. Proposed and adopted this approach in all of Cloud Dataflow and Beam. The impact is that Beam can do a lot of things that users really like doing, but can't do with Spark or Flink.
• https://s.apache.org/splittable-do-fn - created and deployed a new primitive in addition to Map and Reduce, that closes the last remaining gap between batch and streaming data processing - the gap in APIs for data ingestion - allowing Beam to have libraries of data connectors that are completely agnostic to whether they are used with a finite or infinite amount of data. It's also the final missing link in the previous bullet being applicable in all cases.
• internal performance debugging tools - created lightweight APIs and visualizations of what's going on in a data processing job, that gave rise to an ecosystem of hundreds of features and dozens of projects built on top of them inside Google, much more powerful than even commercial offerings for Spark/Flink/Hadoop.

Sep 2010 – Dec 2010 · 3 mos

• Taught a course on functional programming: http://logic.pdmi.ras.ru/csclub/courses/fprog

Jun 2010 – Jan 2013 · 2 yrs 7 mos

• Custom-built low-latency distributed computing infrastructure.

Jul 2009 – Jan 2011 · 1 yr 6 mos

• Russia's only functional programming journal. The closest worldwide analogue is Monad.Reader.
• Wrote and reviewed articles, engaged authors, etc.

Feb 2009 – Jul 2010 · 1 yr 5 mos

• Based on SICP, but concentrating on teaching students to use the advantages of functional style in everyday programming. Students rated the course 5.5 out of 5.
• Led Master's thesis for two students - http://code.google.com/p/valz (a framework for real-time large-scale distributed data aggregation) and http://code.google.com/p/logophagus (a tool for working with log files of unlimited size).

Feb 2007 – May 2010 · 3 yrs 3 mos

• Web IR; DOS attack detection/prevention; general-purpose projects like build script optimization, etc.

Jan 2006 – Jan 2007 · 1 yr

• A simple Windows modem driver, a Java log viewer.