As a part of the Experimentation Platform Team:* Improved a statistical engine with industry-standard methods: Sequential Testing, Outlier Filtering, and CUPED that increased sensitivity by up to 30%. * Developed and maintained an internal statistical library for Data Scientists and Analysts across the company.* Developed and introduced new metric taxonomy and centralized catalog to improve transparency and rigor.* Took an active part in research and developing Multi-Armed… Show more As a part of the Experimentation Platform Team:* Improved a statistical engine with industry-standard methods: Sequential Testing, Outlier Filtering, and CUPED that increased sensitivity by up to 30%. * Developed and maintained an internal statistical library for Data Scientists and Analysts across the company.* Developed and introduced new metric taxonomy and centralized catalog to improve transparency and rigor.* Took an active part in research and developing Multi-Armed Bandits in the experimentation domain. Show less

* Managed team of four people working on DS (recommendation system, uplift modeling, scoring models) * Developed and rolled into production on AWS next services: - Near real-time recommendation system for content collections using multi-armed bandit algorithms, which increased average watched time per visitor by 9% - Service which personalizes any editor’s content collection based on existing recommendation system - Service which provides match score for each user… Show more * Managed team of four people working on DS (recommendation system, uplift modeling, scoring models) * Developed and rolled into production on AWS next services: - Near real-time recommendation system for content collections using multi-armed bandit algorithms, which increased average watched time per visitor by 9% - Service which personalizes any editor’s content collection based on existing recommendation system - Service which provides match score for each user and title pair Show less

* Took part in developing recommender system which increased average revenue per visitor by 6%, reduced the content selection time by 20%.* Developed re-ranking model over Matrix Factorization based on Gradient Tree Boosting * Made analytics around recommendation from scratch: dashboards, reports, issue alerts, A/B tests* Developed and rolled into production revenue forecasting service using Airflow* Developed uplift model which selects relevant offer for each customer

* Made reports and dashboards (Splunk, Python)* Made A/B tests result reports * Extracted insights from user behavior data using common unsupervised learning algorithms