In January 2025, Acherman refactored the CMAP training pipeline to support configurable model creation, focusing on modularity and reproducibility. By updating the create_model function in the dsi-clinic/CMAP repository to accept explicit configuration parameters, and modifying one_trial to pass these arguments, Acherman decoupled configuration from global state. This approach, implemented in Python and leveraging deep learning and model training expertise, improved testability and enabled rapid experimentation with different training setups. The work addressed the need for scalable experimentation and reduced the risk of unintended production changes, demonstrating a thoughtful application of refactoring and machine learning best practices.