Michael Tkachuk

- Integrated MLflow into the development workflow for the Data Science department to use. - Deployed and customized JupyterHub as a replacement for Vertex AI to reduce DS compute costs by 70%. - Designed and developed Ray platform to support large-scale model training and serving.

- Automated face 3d printing pipeline to reduce skincare device manufacturing time from 30 to 8 minutes. - Improved video-based face reconstruction accuracy by 300% (from 20 to 5 mm) and met medical standards. - Developed facial analysis and skin aging evaluation AI algorithms in Python at 90% relative accuracy. - Wrote a fully functional wrinkle detection IOS app alone in less than 2 months. - Set up batched model inference and data communication on AWS using AWS Batch, ECS, S3.

- Customized Stable Diffusion to reach 98.9% accuracy in object detection and segmentation models. - Optimized model training on Azure clusters to run more than 40% faster, saving 100k$ monthly. - Deployed segmentation and detection models on devices using Coreml and TFLite to achieve 20% lower latency.

- Conducted research in satellite imagery analysis using Qgis, exploring effective techniques to parse and interpret satellite data to deliver agricultural insights. - Implemented several domain adaptation papers based on features alignment and GANs.

Top 1 in Linear Algebra, Calculus and Discrete Maths in Class 2023