Developed a configurable cache directory feature for MLIP models within the FAIR-Chem/fairchem repository, focusing on enhancing deployment flexibility and reproducibility. The work involved updating core Python APIs, specifically get_predict_unit and get_isolated_atomic_energies, to accept a cache_dir argument, allowing explicit control over model storage and loading locations. This approach addressed user needs for customization across diverse environments and streamlined CI/CD processes. Leveraging skills in machine learning, model management, and software engineering, the implementation reduced operational friction and improved the overall user experience by making model handling more adaptable to different deployment scenarios without introducing breaking changes or complexity.