Worked on expanding deployment capabilities and improving GPU resource management across two repositories. In bytedance-iaas/vllm, contributed to cloud deployment by adding and updating documentation for deploying vLLM on Modal, a serverless computing platform, using Markdown and Python to clarify cross-platform deployment steps. In gpu-mode/discord-cluster-manager, implemented variant-aware GPU scheduling by parsing GPU names to distinguish between A100-80GB and H100 variants, enabling more granular resource allocation and reducing idle GPU time. Demonstrated backend development and cloud computing skills, focusing on technical documentation, serverless architecture, and precise scheduling logic to enhance cluster utilization and streamline onboarding for new deployment targets.