Delivered and documented Python script that fully automates the image segmentation process of the research, saving 100+ hours of manual measurements for the batch of around 10000 images. Achieved a 2x decrease in systematic error of the measurements by optimizing the image segmentation algorithm via Image Masking. Proposed and completed Geometric Image Transformation in Python for the calibration of the ruler from the images instead of manual computing.
Classified JavaScript scripts, with a team of 4 researchers, affecting the performance of the websites into critical and non-critical classes through a custom-made JavaScript Analyzer tool. Reduced overfitting of the Transformer model by Normalizing input dataset with feature scaling method and designing neural network layers.
Rebuilt the front-end for the main "Math Pages" with additional features and mobile responsive design, resulting in UX improvements of users' workflow and improved maintainability (React.js, TypeScript, Lerna.js). Accelerated development process for teammates by 30% via building reusable React.js components for tab-bar, search and filter functionalities. Integrated 10+ new API endpoints to the assigned web pages enabling appropriate user authorization privileges and error handling using React.js monorepo.