Kevin Han

Led end-to-end AI inspection platform development covering data ingestion, preprocessing, model inference, post-processing, and automated reporting pipelines. Designed and deployed high-resolution YOLOv5x detection solutions optimized for small-object recognition in large-scale aerial inspection workflows. Built synthetic data generation pipelines using Stable Diffusion and LoRA fine-tuning to address limited real-world datasets and improve model robustness. Developed high-performance inference services using ONNX and TensorRT in modern C++, achieving millisecond-level latency in production environments. Delivered structured AI outputs (JSON reports and automated PDFs) integrated into downstream business and data systems. Collaborated with cross-functional engineering teams to continuously improve system reliability and deployment workflows.

Redesigned vision processing modules in modern C++, replacing memory-intensive Halcon workflows and reducing peak memory usage by ~60%. Built scalable visualization and data processing pipelines enabling interactive rendering of 50M+ chip units within ~600 ms. Implemented geometric data processing algorithms including ICP registration and RANSAC-based modeling for precision measurement workflows. Supported ongoing system optimization, maintenance, and performance improvements in production environments.

Developed computer vision and data processing modules using OpenCV, Halcon, and deep learning models. Built multithreaded Qt-based applications integrating cameras, sensors, PLCs, and robotic systems into unified inspection platforms. Participated in full software development lifecycle including system integration, testing, deployment, and long-term maintenance.