Worked on enhancing the testing framework for the tensorflow/datasets repository, focusing on shard computation and dataset writing reliability. Leveraged Python and Beam to expand test coverage, particularly around data sharding and file management across multiple split configurations. Refactored tests to ensure both writers were validated before finalizing outputs, which improved the robustness of dataset writing and reduced the risk of regressions. The approach emphasized preventative validation rather than reactive bug fixing, supporting more stable dataset delivery and continuous integration feedback for downstream users. This work deepened the reliability of the data engineering pipeline and strengthened the overall testing process.