Leading the traffic lane detection project. Created state-of-the-art computer vision model for lane detection that handles vehicle occlusion, unpainted markings and road curvature. Integrated the model into production mapmaking pipelines.Leading data annotation effort for the lane detection project. Training and curating the annotators, providing guides and documentation.Analyzing CV frameworks, model implementations, research papers related to automated mapmaking. Advocating the most efficient ML solutions and practices.Providing prototypes and initial implementations of computer vision models for mapmaking objects detection pipelines. Technologies: Python, Pytorch, MMDetection, Kubernetes, Kubeflow, Flyte, Spark.

Solved varied ML and data science problems as a member of small R&D team of a ride-sharing startup operating in highly competitive New York market. Some of the activities: communication with stakeholders, market analysis, goals/problems setting, data collection and processing, analytics, A/B testing, predictive models development (supply/demand, ETA), production microservices development.Iteratively designed, implemented and supported several versions of Juno dynamic ride pricing system. Owned the full lifecycle of the system.Created marketplace simulator to analyze ride distribution patterns. Implemented and compared rider-driver matching algorithms using the simulator.Developed libraries and pipelines for data collection, analysis and model training.Some of the technologies: Python, Pytorch, Keras, scikit-learn, asyncio, AWS cloud platform.

Co-developed core backend microservices for Iron.io serverless computation platform. Projects: - IronMQ - multi-cloud scalable, highly available and performant message queue with built-in persistence, monitoring and SLA configuration (Golang, RocksDB). - IronWorker - hosted containerized job processing platform as a service (Golang, Ruby, MongoDB). Developed integration test suite for Iron.io platform backend.

Created numerical models of fiber optic lasers, amplifier and communication links. Developed fiber optic systems simulation and optimization software (Wolfram Mathematica, Julia, Matlab, C, C++).Scientific interests: nonlinear system dynamics, synergetics, self-organization.

Co-developed high load, fault-tolerant systems for gaming platform with 1M+ concurrent users online. Projects: - Notification center - unified email, SMS and World of Tanks game client notification system (Erlang, RabbitMQ, PostgreSQL). - World of Tanks game web portal (Python, Twisted, Django). - Deployment and monitoring automation for the above-mentioned systems (Zabbix, Fabric, Team City).