• Served as a technical PM, clarifying user needs and overseeing development schedule and progress. • Developed a CNN-based algorithm for classifying abnormal discharge events in power components , achieving a 75% F1 score. Utilized Grad-CAM for model validation to align features with expert insights. • Developed FastAPI backend APIs to integrate ML predictions, enabling user feedback via frontend, enhancing model accuracy through retraining. • Utilizing in-house TSMC MLOps platforms to achieve CI/CD, and to accomplish data, models, parameters, code version management, and model monitoring & retraining.

• Developed wafer defect inspection algorithms, integrating rule-based and deep learning models, achieving an 80% F1 score. • Created a YOLOv5-based (Object Detection) safety monitoring system for construction sites, assessing equipment integrity and identifying unsafe behaviors, with a 95% F1 score in POC tests. • Overcame limited data challenges in early experimental stages using few-shot and active learning, enhancing project progression.