Developed and delivered a new roll operator for tensor manipulation in the FlagOpen/FlagGems repository, enabling efficient data reordering for machine learning workloads. The implementation leveraged Python and Triton to accelerate operations on GPUs, with careful handling of empty tensors to ensure robustness. Performance optimization was a key focus, resulting in faster tensor operations. Additionally, the developer enhanced the unit testing framework to prevent conflicts during mixed test runs, which improved continuous integration reliability and reduced flaky tests. This work contributed to the FlagGems Operator Development Competition and improved both maintainability and test coverage for future operator development efforts.