Segmentation tasks [Pytorch-lightning, PyTorch, Hydra, OmegaConf, OpenCV, Numpy etc]- Discovered and built time consistent and robust neural model for smooth sport video segmentation for core technology using PyTorch-Lightning and Hydra structured configs- Integrated best trained model into the real-time segmentation inference pipeline (approx. 25 fps) with several functions, such as automatic objects tracking, keyboard shortcuts for zooming and cropping modes- Implemented Hydra structured configs for faster and handy model tuning and experimenting- Adapted RobustVideoMatting model architecture for multiclass task using PyTorch-Lightning with Hydra configs as one of re-searched approaches- Conducted lots of experiments with segmentation models training with IoU test metric from 0.97 up to 0.98- Achieved TV quality streaming image by cameras and lenses search, selection and testing to help broadcast support team- Created several tools to automate data selection and speed up routine ETL operations 3 times faster than manually by human- Increased code base documentation coverage to facilitate adaptation of team members

Data Science, Machine Learning, Deep Learning (Computer Vision)- Developed an object detection and counting system with 98% accuracy to reduce audit department time consumption by 50%- Developed disease symptoms classifier on audio to increase animals survivability- Prototyped labels information recognition on fast moving factory conveyor- Configured label-studio annotation tools and taught team to use it to increase labeling ability in high demand periods for the most ef-ficient datasets creation and model training- Analyzed 2 factory’s data pipelines to find difference in data operations and management systems between them- Created numerous amount of datasets from scratch by exploring and analyzing large amount of visual and audio data

Data Analysis, statistics, geological modeling- created two 3D mathematical-geological models with unconventional modeling techniques- automated modeling workflow with built in functionality and Python scripts- analyzed data of 300 wells to find and model structures in the most efficient way- developed several Python scripts to automate ETL operations, automate well data analysis and creation of geo-statistics reports