During this period, Carlo E. Piross worked on backend reliability and tensor operation efficiency across two major repositories. For pytorch/pytorch, he addressed tensor precision compatibility in the MAIA backend, implementing Python-based fixes to ensure correct tensor handling when double-precision support was unavailable, which improved stability for MAIA users. In NVIDIA/TransformerEngine, he enhanced NVFP4 TensorStorage cache management by introducing conditional cache resets in the weight workspace, leveraging data engineering and machine learning expertise. This approach optimized quantized tensor operations and reduced cache misses, contributing to more reliable and performant model pipelines. His work demonstrated depth in backend and tensor manipulation.