Worked on improving the reliability of model evaluation metrics in the TorchRec repository by addressing a bug in the GAUC calculation. Focused on ensuring that weight handling in the GAUC metric correctly incorporated weights only when labels were present, and ignored them otherwise, thereby preventing misleading evaluation results. Utilized Python and applied data science and machine learning expertise to enhance the accuracy and trustworthiness of GAUC-based model selection processes. This targeted fix strengthened the integrity of evaluation pipelines, supporting more informed deployment decisions and demonstrating a careful approach to metric correctness and code quality within the context of large-scale recommendation systems.