Developed dynamic LoRA adapter loading for the vllm-project/vllm repository, enabling the server to load and configure LoRA adapters at runtime from remote sources. This approach reduced operational overhead by allowing updates without full redeployment, leveraging Python, asynchronous programming, and backend development skills. Additionally, addressed model stability in the OpenPipe/ART repository by fixing a configuration issue, setting the MCP example temperature parameter to 1.0 for more consistent and reproducible model outputs. The work demonstrated attention to configuration management, testing, and clear commit practices, focusing on targeted improvements that enhance model management flexibility and experimental reliability across both projects.