Worked on stabilizing data-loading correctness in the Lightning-AI/pytorch-lightning repository by addressing a bug related to shuffle detection when using batch_sampler. The solution involved inspecting the batch_sampler’s internal sampler to accurately determine shuffle status, with a fallback to the DataLoader’s sampler when necessary. This approach improved the reliability of data-loading, especially for users customizing batch_sampler configurations. Added comprehensive validation tests, including scenarios with batch_size set to one, to ensure robust behavior across different DataLoader setups. Utilized Python and focused on data loading, testing, and utility development, ultimately enhancing training reliability and reducing silent misconfigurations in production machine learning models.