Roman is a Data Scientist with over 8 years experience. His areas of expertise include NLP, Big Data and Image Processing with CNNs. He has worked on a range of projects, mostly related to digital marketing.
St. Petersburg Polytechnic University 1991 - 1996
MS in Computer Science
2018 - Present
Freelance Senior Data Scientist
- Built predictive models for Li-ion and superconductors duty cycles with recurrent neural networks.
- Carried out feasibility study for SSD and YOLO deep learning computer vision models.
- Constructed massive geospatial data processing pipeline for employee vs customer classification.
- Facilitated data engineering aspects of propensity models for major pharmaceutical retailer.
NYU Tandon School of Engineering 2018 - 2017
Data Science Consultant
- Developed course documents, code samples, and use cases for "Machine Learning in Finance" course.
- Built a stock price prediction model from fundamental data using a recurrent neural network.
StepChange 2016 - 2018
- Directed development of a user segmentation pipeline based on mobile historical records.
- Built 95% accurate gesture recognition pipeline for wearable electronics.
- Constructed a chatbot ecosystem intended for easy customization and customer data integration.
- Analyzed a major European mobile network's profits/losses based on users' historical data.
- Improved the churn model performance by 25% using mobile network social data.
Radiumone 2012 - 2016
- Measured effectiveness of mobile ad campaigns using Hadoop, Hive and Python.
- Built targeting segments for a major US airline with geo location.
- Developed deep learning high cost media filtering system that reduced company's media expenses by 5%.
- Architectured a distributed real time GPU-powered time series database.
- Implemented a set of tools for processing and visualizing large geographical datasets (used C++, Cuda).
- Developed predictive content classification pipeline that reduced classification costs by 90%.
- Developed a model for social data sharing increasing performance by 100%+ for selected audiences.
Roman S. is now available for hire