During October 2025, Fryu enhanced model deployment performance for the kaito-project/kaito repository by implementing NVMe local caching for model files. This work involved architectural changes to introduce a caching layer that stores model files on high-speed NVMe storage, reducing load times and inference startup latency. Fryu developed cache management and prefetching strategies, integrated benchmarking to quantify performance improvements, and updated project documentation in Markdown to reflect the new architecture. The focus on performance optimization and clear documentation demonstrates a methodical approach, delivering measurable business value by accelerating deployment cycles and improving runtime responsiveness for machine learning model deployments.