Valentin -

Machine learning engineer at one of the largest e-scooter sharing companies in Russia. Responsible for demand prediction, scooter allocation automation, and user behavior analysis. Developed a demand forecasting model by district and time of day with aggregated weather, time, and behavioral features. The model improved scooter distribution and reduced idle losses by 14%. Built a classifier for “double rides” using sensor data (acceleration, speed, tilt), allowing automatic detection of rides with two people and reducing such violations by 25%. Optimized night charging routes by clustering service zones and applying CVRP (Capacitated Vehicle Routing Problem), which decreased mileage by 18% while maintaining coverage. Analyzed scooter usage patterns and segmented users based on activity time, ride length, frequency, and geography. The insights were used to build an LTV segmentation model.

Worked on data science projects for operations, personalization, and marketing in the dark kitchen business unit. Trained a daily/hourly order forecasting model with weather and promo features. Achieved 5.9% MAPE, improving supply planning and scheduling. Built behavioral segmentation using clustering, identifying 5 stable user groups. This enabled targeted push campaigns and boosted repeat order conversion by 18%. Developed churn prediction model, increasing accuracy by 12% compared to previous heuristics. Implemented a recommendation system for promotions based on purchase history and temporal patterns, increasing average check by 7%.