Worked on enhancing the robustness and reliability of the multimodal generation server in the kvcache-ai/sglang repository, focusing on backend development with Python and ZMQ. Addressed a critical bug by refining ZMQ binding and improving error handling during model loading, which reduced runtime failures and improved server stability. Adjusted server argument management to streamline deployment and ensure smoother integration with FastWan, directly supporting production workloads. All changes were integrated with clear commit traceability, facilitating collaboration and future maintenance. The work demonstrated a methodical approach to backend reliability, emphasizing error management and compatibility in a fast-moving, production-oriented environment.