Worked on the vllm-project/aibrix and yhyang201/sglang repositories, focusing on backend stability and documentation clarity. Addressed a memory leak in the Radix Cache LRU by clearing self pointers on evicted nodes, preventing reference cycles and improving long-term service reliability. Enhanced maintainability by delivering a concise, well-documented fix in Python, emphasizing robust memory management and data structure integrity. In parallel, improved the heterogeneous GPU features guide by removing duplicated lines, streamlining onboarding and reducing confusion for new contributors. Demonstrated attention to detail in both backend development and documentation, leveraging skills in Python, RST, and technical writing to support project quality.