Experience

Ireckonu

• Netherlands
• IT Services and IT Consulting
• 1 - 100 Employee

• Data Scientist

  • Sep 2023 - Present



EPAM Systems

• United States
• IT Services and IT Consulting
• 700 & Above Employee

• Computer Vision Engineer

  • Jun 2022 - May 2023

  • Developed a unified image preprocessing tool that significantly improved overall efficiency and reduced development time, benefiting the integration of new models. • Created a user-friendly UI using Streamlit for visualization and configuration of preprocessing settings, simplifying the process for users and making it more accessible. • Designed and implemented a comprehensive training and testing pipeline for semantic segmentation models, specifically addressing the business task of fruit quality control. Overcame the challenge of a small and unbalanced dataset by conducting extensive experimentation with various loss functions, resulting in 20% improvement in accuracy. • Utilized advanced computer vision techniques to fine-tune a dance generation model. Produced a captivating demo video highlighting the model's potential, featuring retargeted motions applied to a 3D character using Blender. • Leveraged Stable Diffusion models to generate innovative sneaker designs, implementing textual inversion techniques. Demonstrated the viability of the approach through a compelling proof of concept, showcasing the potential for creating unique and visually appealing sneaker designs. Show less



Sberbank

• Russian Federation
• 700 & Above Employee

• Middle Computer Vision Engineer

  • Jun 2021 - Mar 2022

  • Developed a pipeline for training and testing object detection model using Pytorch and a chatbot for model usage • Designed a synthetic dataset for object detection which increased mean average precision (mAP@0.5) by 15% (final mAP@0.5 = 0.97) • Developed train pipeline with automated hyperparameter tuning and iterative pruning for classification model which increased inference speed by 30% • Deployed model into NVIDIA DeepStream SDK pipeline for inference on NVIDIA Jetson • Developed a pipeline for training and testing object detection model using Pytorch and a chatbot for model usage • Designed a synthetic dataset for object detection which increased mean average precision (mAP@0.5) by 15% (final mAP@0.5 = 0.97) • Developed train pipeline with automated hyperparameter tuning and iterative pruning for classification model which increased inference speed by 30% • Deployed model into NVIDIA DeepStream SDK pipeline for inference on NVIDIA Jetson



Сenter 2m

• Moscow, Moscow City, Russia

• Junior Computer Vision Data Analyst

  • Mar 2020 - May 2021

  • Designed and executed experiments to evaluate the performance of various object detection models for a task of recognizing old off pattern markings on pipes. Created image preprocessing algorithm which increased mean average precision (mAP@0.5) by 10%. • Presented research results and developed pipeline with inference and postprocessing on tender and won. • Developed segmentation postprocessing with image filtering on the task of remote weighing of animals which increased business metric by 7% Show less