Jefin Paul

Currently working on R&D projects to modernize the existing cosmetic inspection module for pharmaceutical glass manufacturing. Conducted extensive research on classical and algorithmic computer vision techniques using OpenCV, Eigen and other libraries. Designed and implemented advanced image-processing algorithms in C++ for dimensional inspection of pharmaceutical glass products. Developed new custom measurement methods, optimized and resolved issues in existing measurements to enhance accuracy and reliability. Refactored legacy codebase to modern C++ standards, improving readability and maintainability and enabling seamless integration of new algorithms into production.

Led an advanced R&D initiative integrating AI-driven solutions into plasma technology for precision etching, with successful implementation during the sensing phase. Developed an ensemble-based regression framework leveraging boosting models (XGBoost, AdaBoost, etc.) for predictive accuracy. Applied nested cross-validation and grid search for hyperparameter optimization. Deployed a prototype ML workflow on AWS SageMaker to validate model performance and assess prediction consistency between offline training and production environments.

Conducted advanced object detection experiments using deep learning algorithms like YOLO v6 with Mobilenet backbone, enhancing defect identification in glass products. Developed a custom heuristic algorithm in Python to solve the Multichoice-Knapsack problem, optimizing dataset partitioning and boosting model training efficiency. Increased the IoU metric by 17% and reduced false negatives, addressing critical issues in defect detection. Collaborated with cross-functional teams to translate complex technical insights into actionable business strategies.