During November 2024, Bhargav Buddharaju enhanced the robustness of the MambaStack component in the swiss-ai/Megatron-LM repository by refactoring its layer allocation logic to explicitly handle invalid layer types. Using Python and leveraging deep learning model architecture expertise, Bhargav introduced error handling that raises a ValueError when misconfigurations occur, preventing runtime failures during model construction. He also developed targeted unit tests to validate this behavior, ensuring future maintainability and reliability. This work focused on improving error detection and stability in model deployment pipelines, demonstrating a thoughtful approach to testing and error handling within complex deep learning systems.