During their time on the BerriAI/litellm repository, Peipei developed image caching control options for tool results, focusing on explicit cache management for both image URLs and base64-encoded images. They enhanced the backend logic in Python to support configurable caching behavior, addressing the challenge of handling ephemeral image content and reducing the risk of stale data in tool outputs. Their work included updating the API layer and expanding automated tests to ensure reliability and performance improvements. By refining image processing and cache control, Peipei contributed a targeted feature that deepened the project’s backend robustness and improved image result consistency.