During September 2025, Vandrei developed a GPU Device Metrics Query Utility for the ROCm/pytorch repository, focusing on structured benchmarking and performance reporting for Nvidia GPUs. The utility enables querying of device hardware limits, such as theoretical FLOPs and memory bandwidth, and introduces methods to compute performance metrics based on device capabilities. Vandrei’s work established a benchmarking-ready pipeline and integrated reporting hooks, supporting data-driven optimization and cross-repository performance dashboards. The project was implemented in Python, leveraging CUDA and GPU programming expertise. The depth of the work lies in enabling consistent, automated performance visibility across CUDA-enabled environments without addressing bug fixes.