Contributed to bytedance/deer-flow by delivering two core improvements focused on AI integration and backend development using Python. Developed a personalized memory prompt system that injects long-term user context into system prompts and persists this data for enhanced AI response relevance, supported by comprehensive regression testing. Improved middleware robustness by addressing TypeError issues in LoopDetectionMiddleware when handling list-type AIMessage content, introducing a static method for safer text appending and expanding unit test coverage. Emphasized middleware design and thorough testing practices, resulting in more reliable personalization and safer multi-provider deployments while reducing support incidents and improving maintainability across the codebase.