Worked on the microsoft/onnxruntime-genai repository to deliver quantized model layer support for q_norm and k_norm, addressing the need for accurate quantized attention in newer large language models. Implemented these layers as Tensor modules in Python, ensuring that weights and biases are correctly mapped during model loading for precise quantization behavior. This approach improved compatibility with the latest quantized LLMs, reducing deployment risk and enhancing both inference speed and memory efficiency for generative AI workloads. The work demonstrated expertise in deep learning, model optimization, and quantization, focusing on robust engineering solutions for advanced machine learning infrastructure.