Products Catalog.Created a complete table of products sold at X5 grocery stores with the full set of their characteristics. Increased the number of product categories by 2.8 times. Increased data completeness by 1.7 times. Challenges: finding and merging different existing data sources, extensive data cleaning, and making predictions to fill empty places using the NLP algorithm. Result: granulated detailed reports can be made about product group sales.LLMs.Prototyped the RAG pipeline for creating a question‑answering system with the saiga‑mistral‑7b model. Worked on improving document reranker based on BERT cross‑encoder fine‑tuning.A/B Testing.Designed and evaluated experiments that show whether or not some intervention on grocery stores has a statistically significant effect on some sales metric. 2 successful initiatives were implemented, 7 initiatives with no effect were rejected.Reports Automation.Created a library that simplified and automated additional statistical analysis of the sales data. Reports that took about 4 days to create are now created within 4 hours using 10 lines of code.

Worked on a problem of suggesting short email replies.Created a deeplearning model that predicted the type of email.

Used ML in various NLP problems, including fraud detection and named entity recognition. Implemented the Black-Derman-Toy model of short-rate evolution.

Prototyped a recommender system for bank clients using neural networks (DSSM architecture on keras).Based on previous transactions of a client, the system had to predict which restaurant the client would prefer