• Established R&D team and secured necessary hardware resources.• Revamped the recommendation system strategy, and organized the backlog, led a team of fifteen people to launch MVPs of personalization and ranking algorithms.• Collaborated with the product analytics team to revise metrics and develop essential dashboards, launched a rework of the internal AB-testing tool and conducted several AB-tests.• Managed the implementation of an internal speech-to-text solution, including dataset collection, quality assessment, model selection and resource allocation.• Launched several video and comment analysis models to optimize the moderation pipeline.

During my work at Intema - an AI-first venture vehicle focused on teams of outstanding engineering capability building cutting-edge deeptech products - three main areas of ownership could be distinguished:• Acceleration program: as a part of the C-level team I secured the foundation of our acceleration program roadmap through market research on top players (YC, TechStars, Hyper, etc) and acquiring experts and mentors from my extensive technical network• Startup tech review: as a tech expert I reviewed provided documents, interviewed teams regarding the developed solution and took part in the expert committee in line with CIO and investment managers. 15 companies were chosen for acceleration out of 1000+ total. For the evaluation of the startups an internal assessment system was developed focused on an in-depth understanding of underlying AI tech and the market fit of the product. That process was so efficient that 13 companies out of 15 selected were given outstanding scores at final tech due diligence.• Intema platform: as a product owner I initiated the development of a platform, connecting startups with investors and corporations. I provided the team with the roadmap to execute based on market research on similar solutions (AngelList, SberUnity, AngelHub) and 10+ custdev interviews

I worked on OCR/RPA system powered by Deep Learning and Humans-in-the-Loop, enabling people across the globe to improve document recognition quality in real time. I ensured that company’s tech is advanced enough to fuel it’s exploding growth by:• Leading the multi-functional team of 15 people to transform the product into enterprise-ready solution through building reliable multi-GPU architecture for scaling complex DL pipelines• Introducing a data-driven method of team capacity estimation and ICE-inspired task prioritization, which raised team velocity by 80% to date• Inspiring the team to integrate advanced deep learning inference acceleration technologies, speeding up our pipelines by an order of magnitude• Defining the lacking skillsets within the team and hiring multiple developers to fulfill the demand• Implementing a system to boost crowdworkers performance through allowing bizdev team to configure descriptions, rules and suggestions for tasks at Y.Toloka. Resulted in up to 99.7% accuracy, 30% lower cost and 40% speed-up• Building a framework to automatically define document structure, drastically speeding up the process of enabling recognition for novel document types

At iPavlov, an internationally renowned lab dedicated to advancing state of the art in NLP AI, I led an engineering team on a mission to integrate conversational AI into customer support workflow at Sberbank (the largest Russian bank). Project results:- Stateless NLP-service, providing intent classification and NER, integrated with bank CRM through Kafka, built in Python, currently being tested in production by support agents.The second project I am working on as a sole backend developer revolves around building a SaaS solution to enable external clients to use deep learning models created by iPavlov. Current project state:- Horizontally scalable cloud service that balances load between multiple computation-heavy ML models. Built in Go, RabbitMQ, Python, Clickhouse, PostgreSQL, Docker. Used by multiple clients in production.

As a software engineer at MySky I was responsible for speeding up private jet financial accounting system. I made that system maintainable and fast through:- Porting the codebase to Python3 from Python2 - Implementing Cap'n'Proto serialization protocol to speed up interactions with the front-end - Optimizing complex SQL queries on the business logic layer - Implementing a modular system to aggregate financial metrics over an arbitrary time frame

I enabled the analytics team to model public transport of Moscow by building a set of tools to aggregate geo-spatial big data into daily reports on usage and transport load based on detailed logs of citizens movement.

I took part in all the stages of integration of the clinic management solution into one of the largest federal hospitals in Russia, from collecting requirements for additional features to implementing those features in Delphi and VBS.