Jinghua worked on the vllm-project/aibrix repository, focusing on improving the reliability of benchmarking for Chain-of-Thought LLMs. Using Python and leveraging skills in benchmarking and LLM integration, Jinghua addressed a critical issue in Time To First Token (TTFT) measurement during streaming. By refining the logic to accurately capture reasoning content even when it was not immediately available, Jinghua enhanced the accuracy and trustworthiness of benchmarking data. This bug fix improved the quality of TTFT metrics, enabling more reliable model comparisons and supporting data-driven evaluation. The work demonstrated careful debugging and thoughtful instrumentation within a complex benchmarking pipeline.