Developed a hot-reload capability for the kvcache-ai/sglang repository, enabling in-place model weight updates from disk within the SGLang-Diffusion Engine. This feature allowed model weights to be updated at runtime without requiring server restarts, reducing deployment downtime and supporting faster iteration cycles during machine learning experimentation. The implementation leveraged Python and FastAPI for backend development, focusing on runtime state management and disk-based weight loading. By enhancing model management workflows and enabling zero-downtime updates, the work aligned with goals of reliability and scalability. Collaboration included cross-functional code review and integration, demonstrating effective teamwork in a machine learning engineering context.