Worked on the pytorch/torchrec repository to address a bug in the NDCG metric calculation for multitask recommender system scenarios. Focused on improving the accuracy and reliability of offline evaluation by fixing the NDCG implementation, which reduces the risk of misranking in complex multitask pipelines. Developed and integrated comprehensive unit tests using Python and pytest to validate the metric across multiple tasks, ensuring robust test coverage. Leveraged a Git-based workflow and continuous integration to maintain code quality. This work enhanced the dependability of TorchRec’s evaluation tools for users working with machine learning and data science in recommender system contexts.