Worked on enhancing test coverage and reliability for GPU communication libraries within the ROCm ecosystem, focusing on the ROCm/rccl and ROCm/rocm-systems repositories. Developed comprehensive unit tests in C++ and CUDA to validate NCCL and RCCL device association, communicator error handling, and buffer registration flows across single and multi-device scenarios. Improved robustness by introducing checks for invalid communicator handles and refining debug message ordering, which led to clearer test output and more predictable error reporting. Collaborated across repositories to align test infrastructure with broader CI goals, enabling earlier detection of regressions and supporting safer refactoring of registration logic.