Worked on the sglang repository to deliver a weight-loading enhancement focused on improving model deployment workflows. Developed the ModelExpress package using Python and gRPC, introducing dynamic seed discovery and configurable backends to support robust and flexible weight loading across diverse environments. The approach involved refactoring the loading process to delegate responsibilities to the new package, effectively decoupling concerns and simplifying ongoing maintenance. By abstracting backend logic and supporting multiple configurations, the work enabled more reliable deployments and easier experimentation. This modular loading path strengthened deployment confidence and maintainability, reflecting a thoughtful application of backend development and model deployment skills.