Worked on preparing the livepeer/ai-worker repository for StreamDiffusion deployment and ensuring compatibility with the 5090 GPU. Focused on updating Dockerfiles and managing dependencies to align with StreamDiffusion requirements and support future GPU generations. Utilized Docker and Shell scripting to standardize environment configurations, explicitly pinning PyTorch, torchvision, torchaudio, and Triton versions for stability across toolchains. Addressed deployment reliability by aligning base images and CUDA versions with newer PyTorch releases, enabling maintainable and robust deployments. The work demonstrated depth in CI/CD and DevOps practices, emphasizing containerization and hardware compatibility for evolving AI pipelines within the repository’s ecosystem.