Anton Andreitsev Email & Phone Number

Ranking and Personalisation Team

Moscow, Moscow City, Russia

Lecturer and seminarian on Introduction to machine learning course at Economics Faculty of MSU.

Moscow, Moscow City, Russia

1) Investment propensity models (LAL (Look alike), Response and Uplift models)2) Investments customers churn models3) Investment products recommendation models4) Expected operational profit forecasting modelsI also lead a (3 people) team of data scientists. My team is responsible for the investments block, and customer's wage block (determine customer's wage, his propensity to churn from wage project, elasticity to different percent rates/tariffs etc.)

Moscow, Moscow City, Russia

1) Development of daily demand forecasting model for KFC restaurants in Russia. Result: Model outperformed current KFC model on their test period. 2) RnD in field of model interpretability for KFC demand forecasting model. We were trying to decompose gradient boosting models down to some interpretable components. Result: Managed to extract seasonal and promo components from the forecast. This decomposition is needed to reinforce credibility to model's outputs.3) Forecasting… Show more 1) Development of daily demand forecasting model for KFC restaurants in Russia. Result: Model outperformed current KFC model on their test period. 2) RnD in field of model interpretability for KFC demand forecasting model. We were trying to decompose gradient boosting models down to some interpretable components. Result: Managed to extract seasonal and promo components from the forecast. This decomposition is needed to reinforce credibility to model's outputs.3) Forecasting motor oil blends characteristics for Gazprom. The task was, based on the blend parameters (proportitions, characteristics of base oils, additives, etc) predict multiple characteristics of resulting blend (such as viscosity, pour point, css, etc.).Result: The resulting model managed to achieve target metrics (metrics of Gazprom) on most of the target characteristics Show less

Moscow, Moscow City, Russia

1) Improving model of symptoms recognition. Replacing regexp-bases model with neural net-bases model. Result: model inference acceleration from 8 sec. on "standart medical card" up to faster than 1 sec. with f1-score 99%.2) Developing Eva-gynecological bot (@Eva_Gynecologist_bot - in telegram). There were 2 main tasks: 1) Get to know all the topics, that users ask Eva (Texts clustering, Topic modelling: mainly LDA). 2) Queries classification. User asks Eva some question - we need to… Show more 1) Improving model of symptoms recognition. Replacing regexp-bases model with neural net-bases model. Result: model inference acceleration from 8 sec. on "standart medical card" up to faster than 1 sec. with f1-score 99%.2) Developing Eva-gynecological bot (@Eva_Gynecologist_bot - in telegram). There were 2 main tasks: 1) Get to know all the topics, that users ask Eva (Texts clustering, Topic modelling: mainly LDA). 2) Queries classification. User asks Eva some question - we need to classify that question to some leaf in tree of contents (details could be found here: https://docs.google.com/presentation/d/17XxDfRY8sLQ3uDiGJC1ZvMMAKADJ6BJXKwcTToYrnXM/edit?usp=sharing (it's in Russian though)), so it's a hierarchical classification problem (Tf-idf + classical ml methods, nn-models).Result: Developed hierarchical classification model (about 750 classes) with quality about 80% f1-score. Show less

Master'S Degree, Mathematics And Computer Science

CMC Faculty

Bachelor'S Degree, Economics