- Increased profitability of order by 2,5 % by improving restaurant recommendation system- Created ranking simulation framework which enables to test models offline- Increased business metrics sensitivity by 30% by applying variance reduction technique

Optical character recognition team- Implemented a model based on visual and language transformers which improved hard cases recognition accuracy by 9%- Improved manuscript recognition quality for archive search service by 11%- Trained word coordinates detection model over OCR transformer which increased words boxes IOU by 14%- Researched and applied unsupervised training techniques for both language and visual transformers- Deployed and supported the text recognition service with more than 1mln users per day

SberData team- Time series analytics and modeling with classical and deep learning methods- Application of Wavelet transformation and computer vision methods tomultistep forecasting- Text attributes tokenization and classification- Model inference, ML service deployment

- Financial statements analysis- Hypothesis testing- Scoring model development