During November 2025, Zhaozuy Zhao developed dynamic LoRA adapter management for the IBM/vllm repository, focusing on scalable, SageMaker-ready customization of the vLLM OpenAI API server. Leveraging Python, FastAPI, and backend development skills, Zhao implemented dynamic loading and unloading of LoRA adapters to support stateful session management and multi-tenant model customization. The work included introducing new API routes for adapter requests and middleware to enhance request processing and enable on-demand model behavior adjustments within SageMaker. This feature addressed the need for flexible, rapid iteration cycles in machine learning workflows, demonstrating depth in API and backend engineering for production environments.