Worked on the NVIDIA/NeMo repository to enhance reliability and logging during checkpoint loading by addressing a critical bug in the OneLogger integration. Focused on Python development, the work involved debugging and dependency management to eliminate duplicated start and end notifications, remove redundant logger calls, and streamline initialization by discarding an unnecessary error-handling strategy. Updated the one_logger dependency to maintain compatibility with recent changes, ensuring consistent and reliable checkpointing notifications. These improvements reduced log noise and technical debt, resulting in cleaner logs and more stable training workflows. Emphasized maintainability and testing throughout the process to support production-ready machine learning pipelines.