Worked on the pytorch/torchrec repository to address reliability issues in model input handling tests, focusing on stabilizing continuous integration processes rather than delivering new features. The main contribution involved reverting a previous fix for a flaky test related to input_jkt.weight data types, restoring a stable baseline while ongoing issues were investigated. This approach emphasized careful debugging, rollback strategies, and clear commit traceability. Leveraged Python for both data handling and machine learning workflows, with a strong emphasis on testing practices. The work laid a safer foundation for future improvements in input data type handling and contributed to more reliable test infrastructure.