Worked on the Blaizzy/mlx-vlm repository to enhance server runtime stability and efficiency by addressing a memory leak related to Uvicorn’s auto-reload feature. The solution involved disabling auto-reload by default and introducing a command-line --reload flag for development use, which clarified the distinction between production and development environments. This change reduced memory usage and minimized unnecessary server restarts, directly improving production reliability. The work required backend development skills, particularly in Python, with a focus on CLI argument parsing and server configuration. The update did not introduce new features but delivered a targeted bug fix that improved operational performance.