Developed GPU barrier synchronization for the Mooncake backend, enabling reliable cross-device coordination for multi-GPU workloads. The implementation introduced logic to detect whether operations run on CPU or GPU, dispatching the appropriate task accordingly. This approach improved synchronization reliability and laid groundwork for scalable production deployments. The work included expanding unit test coverage to validate both CPU and GPU barrier paths, ensuring robust behavior and preventing regressions. Collaboration was demonstrated through co-authoring and code review on the main repository. The project leveraged C++, Python, and CUDA, focusing on backend development and comprehensive testing to enhance the Mooncake system’s multi-device capabilities.