Worked on the microsoft/ltp-sglang repository to enhance the CI/CD pipeline, focusing on Docker CUDA and ROCm build environments. The approach involved updating base images, refactoring the image tagging strategy to distinguish between release and stable tags based on Git branch and event, and addressing disk space limitations in build runners. Leveraging skills in Docker, Shell scripting, and Python packaging, the work improved build reliability and consistency across deployment environments. Additionally, Python package installation within Dockerfiles was optimized for better version control and build speed, resulting in a more efficient and maintainable release process for the project.