Worked on the sgl-project/mini-sglang repository to address automation reliability in benchmark scripts that utilize the OpenAI API. Focused on backend development using Python, the work involved fixing a bug related to missing credentials during OpenAI client initialization. By implementing robust error handling and supporting environment-variable fallbacks for API keys, the changes ensured that benchmark workflows could proceed without manual intervention. The solution improved asynchronous programming flows by preventing runtime failures and enabling stable, rollback-safe execution paths. This targeted update enhanced the automation pipeline, allowing for more dependable integration of OpenAI-backed features within the existing backend infrastructure.