Worked on enhancing GPU test reliability and build performance across ROCm/tensorflow-upstream, openxla/xla, and Intel-tensorflow/xla repositories. Addressed command buffer tracking issues by refactoring test logic to maintain dual input-argument sets, ensuring accurate device buffer updates and reducing test flakiness. Improved kernel compilation speed by implementing an in-memory and persistent HSACO cache for the amdgpu_backend, leveraging SHA256 keys for efficient lookup and integration. Utilized C++ and compiler design expertise to deliver robust, documentation-friendly code changes, including build dependency adjustments. The work resulted in more stable CI pipelines, faster compilation times, and improved cross-repository test accuracy for GPU programming workflows.