Developed a data-cleaning feature for the metatensor/metatensor repository, focusing on enhancing data integrity and efficiency in analytics and model training workflows. Built the Drop Empty Blocks operation for TensorMap, which programmatically removes blocks with zero-length dimensions across samples, components, or properties. The implementation leveraged Python and incorporated robust unit tests to ensure consistent behavior across both NumPy and PyTorch backends. This work improved storage efficiency and reduced noise in downstream processes by eliminating irrelevant data. The approach aligned with repository quality standards and established a foundation for future backend-agnostic tensor operations, demonstrating depth in cross-backend engineering and testing.