Data Scientist
CurrentResponsibilities and Achievements:1. Internal Solution Development: - Developed Data Observability solutions to track data distribution shifts, currently undergoing PoC with POSCO.2. POSCO AI Smart Help Center: - Led AI model development and consulting projects across various POSCO divisions.2024 Projects: - Developed shape recognition for conveyor top/tail drop in wire rod manufacturing using MVit (Multiscale Vision Transformer), featured as a top… Show more Responsibilities and Achievements:1. Internal Solution Development: - Developed Data Observability solutions to track data distribution shifts, currently undergoing PoC with POSCO.2. POSCO AI Smart Help Center: - Led AI model development and consulting projects across various POSCO divisions.2024 Projects: - Developed shape recognition for conveyor top/tail drop in wire rod manufacturing using MVit (Multiscale Vision Transformer), featured as a top project in POSCO Today. - Enhanced the post-leveling shape prediction model for thick plates using a Vision-Tabular Multi-modal model. - Developed anomaly detection algorithms for cooler fan motors and gearboxes in the sintering plant using time series data. - Assessed welding quality for 1/2PCM welders in cold rolling using instance segmentation (Yolo v9). - Detected foreign objects on raw material belt conveyors in blast furnaces using visual anomaly detection with DRAEM-SSPCAB model.2023 Projects: - Developed AI video recognition systems for operator detection in cold rolling PCM processes using object detection (YOLO v8). - Developed cobble detection systems to prevent operator accidents in wire rod manufacturing using computer vision (OpenCV). - Enhanced rolling time prediction for hot rolling and developed MPC models to reduce rolling pitch using XGBoost, and automated training, incorporating data pipelines built with PySpark, Airflow and Kubernetes. 2022 Projects: - Developed alarm systems for seam folding in STS rolling using instance segmentation (Detectron2). - Implemented safety monitoring systems to predict crane hook detachment in STS steelmaking using keypoint detection (Faster RCNN). - Established systems to improve yield and reduce costs for STS products by predicting optimal slab edge trimming using semantic segmentation (Unet). - Developed profile prediction models for hot rolling to enhance yield and automate ACL trimming using multi-output regression. Show less