Worked on the PaddleX repository to deliver a reliability-focused feature that streamlines deployment on edge hardware. Developed and refined dynamic device detection logic in Python, enabling the system to automatically prefer GPU resources on devices like Jetson and gracefully fall back to CPU when necessary. This approach removed the need for explicit device flags in YAML configuration files, reducing configuration errors and standardizing deployment behavior across face recognition and PP-ShiTuV2 pipelines. Leveraged skills in configuration management and device management to enhance edge deployment readiness, improve GPU utilization, and ensure consistent, maintainable configuration practices throughout the codebase. No bugs were reported.