Worked on the intel/onnxruntime repository to enhance ONNX operator compatibility by extending support for mixed int8 and uint8 input types in both ConvInteger and MatMulInteger operators. Leveraged C++ and deep learning expertise to implement templated runtime dispatch, enabling all 8-bit input combinations and preserving backward compatibility. Addressed ONNX specification gaps by registering missing kernel paths and resolving runtime errors for mixed-sign matrix operations. Developed comprehensive unit tests to validate new type combinations and ensure ONNX-compliance, strengthening CI readiness. The work improved model portability, reliability, and production readiness for ONNX-based neural network inference, with a focus on quantization and matrix operations.