Resume

Work Experience

As a tech lead of the computer vision team,

Designed and led development of a unified recognition pipeline which was employed in 8 services, superseded 4 legacy pipelines, and decreased new model development time by 70%
Spearheaded the efforts to move 20+ production models from TensorFlow 1.X to PyTorch, and on the internal MLOps platform, improving both model reproducibility, maintainability, and quality metrics
Architected a NER-based data extraction pipeline with 3 new libraries expanding automatic data capture on large texts
Managed technical backlog, coordinated team development efforts, and provided technical guidance to 5 team members
Ran regular OCR/CV/NLP seminar to fill the experiments backlog and generate ideas for model improvement; authored 25% of the 2000+ training jobs and managed all 60+ model deployments generated by the team in a year
Drove development of a core dataset (1m+ labeled image crops, 50+ revisions) and related data access libraries; instructed the annotation team

Technologies: Python, PyTorch, numpy, OpenCV, flairNLP, aiohttp, RabbitMQ, S3, PostgreSQL, git, CI/CD, Docker, Kubernetes, LMDB

As a member of the data science team at a real estate marketplace company,

Implemented and deployed recognition service which increased a number of automatically processed applications by 30%
Built 3 computer vision models for the moderation team (fraud detection models, a room type classification model) to reduce manual work
Redesigned document validation service with a custom multi-task network architecture decreasing response time by 60%
Extended and overhauled document classification pipeline with 40 new document types while reducing inference speed by 50%
Developed and maintained watermark insertion library applied to every property image across the marketplace

Technologies: Python, TensorFlow, Keras, numpy, OpenCV, Sanic, PostgreSQL, ClickHouse, git, CI/CD, Kubernetes, Grafana, InfluxDB

First hire with focus on machine learning for a Marketing-as-a-Service startup working with e-commerce businesses. Built

Coupon recommendation system based on clickstream data which increased product revenue by 15%
Purchase prediction service that spanned 500+ product categories and 50+ sites and was used for customer segmentation
Send time optimization model to schedule marketing campaigns, bringing 5-10% increase in CTR in A/B tests on average
Site exit intent detection model which reduced notification volume by 70% without increasing bounce rate

Technologies: Python, SQL, R, AWS, xgboost, scikit-learn, pandas, Vowpal Wabbit, fastText, Vertica, DynamoDB, Kafka, git

Created portfolio return distribution modeling tool based on historical time series data empowering value-at-risk analysis
Engineered trading terminal plugin for real-time portfolio risk assessment to enable automatic company-wide risk alerts
Automated portfolio stress testing simulation to save 4 man-days each month

Technologies: R, tseries, copula, GARCH, Lua, Excel VBA

Specialist degree in Mathematics and System Programming (M.Sc. in Computer Science)