• Improved the efficiency and quality of the Muzero RL algorithm. Increased the score several times on 7/57 Atari games by new policy of replay buffer samples aggregation• Implemented the reanalyzer which updates data in replay buffer to refresh network targets due to them expiration during training process• Integrated Muesli algorithm to Muzero implementation. Increased mean result on Atari domain on over than 10%

• Developed a server handler for Yandex.Taxi, which, as a result of interaction with many other microservices,collects, aggregates, and displays to drivers the reasons why they are not allowed to take orders• Developed several filters that control the matching of drivers and orders according to certain criteria• Learned and implemented a special technology, which allows changing the settings of the filter by zones, countries, and driver tags without re-coding