Developed a server startup preloading feature for the Blaizzy/mlx-vlm repository, enabling machine learning models and adapters to be optionally loaded at launch via new command-line flags. Leveraged Python and FastAPI to implement robust error handling and streamline the initialization process, resulting in faster server readiness and more predictable deployments. Enhanced the command-line interface to support flexible model management, while updating documentation to guide users through the new workflow. Focused on backend development and API integration, this work addressed the need for efficient ML asset management and improved deployment reliability across diverse environments, contributing to smoother operational workflows for ML workloads.