Kirill Mansurov Email and Phone Number

Conducted research in subject of simulation of the exchange as a multiagent system based on agent-based approach.Proposed improved method for stock-exchange implementation by including self-learning agent, which has no fixed strategy, but can adapt to the current environment conditions. These agent was built on reinforcement-learning approach, policy-based method.The resulting new model improves the accuracy of the market description by 3-4% compared to previous papers.Author of… Show more Conducted research in subject of simulation of the exchange as a multiagent system based on agent-based approach.Proposed improved method for stock-exchange implementation by including self-learning agent, which has no fixed strategy, but can adapt to the current environment conditions. These agent was built on reinforcement-learning approach, policy-based method.The resulting new model improves the accuracy of the market description by 3-4% compared to previous papers.Author of the following paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3992289 Show less

Improved and supported a model for news events forecasting. The model estimates and predicts coverage of the events related to city's news. Improved feature generation, added new types of features, created hyperparameter tuning based on bayessian optimization. As a result the improved model is 16% more accurate than the previous one.Developed a new model for automatic assessment level of employees for city labor exchange. The model is based on decision trees gradient boosting method… Show more Improved and supported a model for news events forecasting. The model estimates and predicts coverage of the events related to city's news. Improved feature generation, added new types of features, created hyperparameter tuning based on bayessian optimization. As a result the improved model is 16% more accurate than the previous one.Developed a new model for automatic assessment level of employees for city labor exchange. The model is based on decision trees gradient boosting method with cv's parser and feature extractor. As a result, the labor exchange was able to significantly improve the recruitment of employees, increasing the speed of the resume screening process by 20%, and reducing the number of people involved by 7%.Support current database server. Created a design data storedronts. Designed data storefronts for such government departments, as planning department, logistics, public utilities, etc. Show less

Created a model for estimating the optimal price for paid services using dynamic pricing algorithms based on multi-armed bandits. Created a new model for assessing optimal demand using Thompson sampling, multi-armed bandits, simulating real buyers of services.The model is capable of changing and retraining if it is necessary.The input data consisted of the purchase history of each customer, taking into account the purchase price of the service, the availability of discounts… Show more Created a model for estimating the optimal price for paid services using dynamic pricing algorithms based on multi-armed bandits. Created a new model for assessing optimal demand using Thompson sampling, multi-armed bandits, simulating real buyers of services.The model is capable of changing and retraining if it is necessary.The input data consisted of the purchase history of each customer, taking into account the purchase price of the service, the availability of discounts, the season, etc.This method allowed us to calculate a fairer amount of demand.The results of A/B testing showed the advantage of a new pricing approach.This increased the company's revenue on the auto.ru source by 2% in 2 months compared to same period in previous years. Show less

Pricing stock derivatives.Developed a method for quickly calculating loan fair value using active learning approach. This approach is based on Bayesian optimization. New method made it possible to increase the speed of calculating the fair value by 200% with losses in accuracy by 0.1%

Developed and implemented new approaches for forecasting gaz injection and punping volumes by EU contries through a gas pipeline.The learning rate of new approach has increased by 20%, forecast accuracy has increased by 7%Implemented new forecasting model based on autoregression methods. The forecast was carried out for a month ahead.

Optimized processes for product line control at the factory. Built a new database architecture for storing reports on manufacturing defects. Developed interface in Microsoft PowerBi for monitoring and make changes in database.