Alban Des developed and delivered a targeted bug fix for the pytorch-labs/monarch repository, focusing on the reliability of gradient propagation in deep learning workflows. By addressing the handling of gradient_generation_order when next_functions lacked a GradientEdge, Alban ensured that gradient computation remained correct and stable, preventing downstream errors in model training. The solution was implemented in Python and required a deep understanding of autograd internals and gradient computation mechanics. This work improved the stability of Monarch-powered models in production pipelines, demonstrating Alban’s ability to debug complex issues and contribute robust patches to critical machine learning infrastructure.