Developed and delivered the MemFabric-Hybrid KV Cache Transfer Enhancement for Ascend NPU clusters within the kvcache-ai/sglang repository, focusing on backend development and hardware-accelerator optimization. Upgraded the MemFabric adapter to a hybrid model, enabling more efficient and scalable KV cache transfers for AI workloads running on Ascend NPUs. The work involved careful integration of MemFabric-Hybrid into distributed cache paths, leveraging Python and Shell scripting to ensure robust commit traceability and maintainability. No major bugs were reported during this period, reflecting a stable deployment. Demonstrated proficiency in CI/CD workflows, Docker, and distributed system design throughout the feature’s implementation.