Created user segments based on health conditionsCreated patient stages classification model based on web activity (symptoms, diagnosed, treatment, recovery)Created similar web-sites recommender modelImplemented cross-platform advertising budget allocation algorithmParticipated in development of NLP model (CNN) for tagging of web pages by 7K health conditions

Led the development of a number of data science projects such as
:- “It’s time to buy” (auto-basket)
- Orders forecast for each SKU. Regular period – 20% WAPE, promo period – 30% WAPE
- Price setting per SKU. Accounting for elasticity, competitors prices and KVIs. +2% in category revenue.- Service time prediction (how long it will take a courier to serve a client). -3% to the delivery cost for the same reliability level
- Graph of substitute SKUs. Used to complete an order when the warehouse lacks initially ordered SKU. -10% drop in orders canceled because of an absence of some SKUs.
- Direct marketing personalization. Uplift modeling, personal digests (sets of promo campaigns)
- Warehouse simulation, slotting improvement. 3x drop in meters per pick.Set working processes (communication, documentation, task tracking) and development principles (code versioning, review, testing, deployment)Managed the team of 7 Data Scientists (project management, people management and everything else) and about 10 external consultants (only project management)

Working as an Individual Proprietor for the following companies:- AITrading (aitrading.com): Created an algorithm of automatic detection of trading patterns in real-time. Established robust ETL pipeline- Quantribution (quantribution.io): Created a part of hedge-fund infrastructure responsible for order book bars calculation in real-time (microservices on Python connected via ZeroMQ)

Dynamic pricingChurn predictionBuilt the model that predicts the class in which a user will make an orderAd-hoc analysis for feature development prioritization

Performed clustering analysis on patients and doctorsSolved the problem of optimal allocation of doctors by clinics

Programmed an infrastructure to compare different roll-forward strategies in futures.

Studied ways to improve Viola-Jones cascades learning.