Worked on the openvinotoolkit/openvino repository to deliver a feature expanding CWAI3 pattern matching for improved inference performance. Focused on generalizing the handling of weight scales as constants, the solution introduced support for optional slicing and conversion steps across various data types and operations. This approach leveraged expertise in C++ and graph optimization, with an emphasis on pattern matching and performance optimization techniques. The enhancement resulted in an observed reduction of approximately 240 milliseconds per inference across representative workloads. While the update improved flexibility and speed, it also surfaced a regression in certain operations, which is being actively tracked.