Worked on enabling GPU backend support for the QNN Execution Provider within the ROCm/onnxruntime repository, focusing on delivering GPU execution capabilities for QNN EP models. Leveraged C++ development skills alongside deep learning and GPU programming expertise to integrate the QnnGpu backend, allowing models to utilize GPU acceleration for improved inference performance and broader compatibility. The approach involved aligning new features with the existing ROCm/onnxruntime architecture to ensure minimal risk and maintain system stability. This work laid the foundation for future GPU-accelerated machine learning workflows, emphasizing maintainability and compatibility while expanding the capabilities of the QNN Execution Provider.