Jordan Taylor focused on backend development and system configuration for the UKGovernmentBEIS/inspect_ai repository, addressing local server stability and CUDA device management. Using Python, Jordan refined the handling of command-line arguments, ensuring empty values no longer caused misconfigurations. The work included updating CUDA_VISIBLE_DEVICES logic to adaptively select parallel processing sizes based on available GPUs, which improved device utilization and reduced environment-related failures. By targeting a specific bug in the local server startup process, Jordan enhanced reliability for development and testing environments. This contribution deepened the robustness of local deployment, streamlining onboarding and reducing debugging time for data science workflows.