Feiqiang Sun developed a FlexKV-based cache offloading connector for large-scale LLM inference in the jeejeelee/vllm repository. Leveraging Python and backend development expertise, Feiqiang designed the connector to enable efficient memory management by offloading key-value caches, addressing scalability and resource constraints in production inference workflows. The implementation included a practical example to guide integration and a comprehensive suite of unit tests to ensure correctness and reliability. By maintaining backward compatibility with existing APIs and focusing on distributed systems and cache management, Feiqiang delivered a robust solution that reduces memory pressure and supports scalable, reliable deployment of large language models.