Yaowei Zhou developed a performance optimization for the FullyConnected operation in the google-ai-edge/LiteRT repository, focusing on Intel platforms using OpenCL with CLVK. By implementing Shared Local Memory (SLM) usage in C++, Yaowei enabled the operation to leverage GPU computing resources more efficiently, reducing compute time and increasing inference throughput on edge devices. The technical approach involved integrating SLM into the existing OpenCL kernel, followed by thorough benchmarking and hardware verification to confirm performance gains. This work expanded LiteRT’s OpenCL support for Intel CLVK, demonstrating depth in performance optimization and GPU programming within a production codebase over the month.