Worked on the google/tunix repository to enhance the VLLM installation process by enabling support for CPU-only environments. Addressed installation barriers by updating the installation script, defaulting the backend to torch-cpu instead of requiring TPU hardware. This adjustment broadened deployment options and improved accessibility for users without specialized hardware, streamlining onboarding and production readiness. The work involved DevOps practices and scripting, primarily using bash, to ensure reliable and traceable installation flows. By focusing on compatibility and reducing hardware dependencies, the changes made it easier for a wider user base to install and run VLLM in diverse environments.