Worked on the langchain-ai/langchain repository to enhance streaming observability and user insights by expanding metadata in streaming responses. Developed the ChatZhipuAI Streaming Metadata Enhancement, which added token usage and model name information to both synchronous and asynchronous streaming methods. This update exposed detailed generation_info, enabling developers and product teams to monitor generation costs, model selection, and performance without altering the existing API surface. The implementation, using Python and leveraging backend and full stack development skills, maintained backward compatibility and required minimal changes to deployment, ultimately improving cost visibility and debugging capabilities for streaming workflows in production environments.