Worked on the NVIDIA/NeMo-Skills repository to enhance training checkpoint management and artifact handling for deep learning workflows. Developed a dedicated workflow for saving the last training checkpoint, distinct from checkpoint averaging, to improve reproducibility and operational efficiency. Refactored the checkpoint command generation process, enabling flexible support for both averaging and last-checkpoint saving within model training pipelines. Introduced a Python utility, copy_checkpoint.py, to streamline the copying and organization of training artifacts across runs. Leveraged skills in checkpoint management, machine learning, and model training to deliver a robust solution that provides greater control and flexibility over training artifacts in production environments.