Jingpu contributed to the Intel-tensorflow/tensorflow repository by developing a Unified 2D Convolution-Bias-Activation operator, which fused convolution, bias addition, and activation into a single operation to streamline CNN model execution. Leveraging C++, MLIR, and TensorFlow, Jingpu’s implementation reduced kernel launches and memory bandwidth usage, laying the foundation for further operator fusion and optimization in production pipelines. Additionally, Jingpu addressed reliability in int8 inference by introducing a targeted guard that disables MHLO to TensorFlow convolution conversion for unsupported 8-bit integer inputs, preventing runtime errors. The work demonstrated depth in compiler design and practical problem-solving for machine learning workflows.