Zhao worked on targeted improvements in both documentation and model management within the langchain-ai/langgraph and liguodongiot/transformers repositories. In langgraph, Zhao enhanced the StateGraph API documentation by correcting type hints, improving clarity and reducing integration friction for developers using Python. Later, Zhao contributed to liguodongiot/transformers by enabling FlaxPreTrainedModel to load checkpoints from local subfolders in safetensors format, supporting more flexible and offline model deployment workflows. The work demonstrated a focus on practical developer experience, with careful attention to detail in both documentation and machine learning model deployment, leveraging skills in Python and model management without introducing runtime changes or bug fixes.