Worked on the fastmachinelearning/hls4ml repository to enhance reliability in model deployment pipelines by addressing a critical issue in tensor scale handling. Focused on machine learning optimization and tensor manipulation using Python, the work involved correcting the extraction of the first element from multidimensional scale tensors to ensure it is always treated as a scalar. This fix resolved calculation errors that previously occurred when interfacing with tensors shaped for frameworks like qonnx, such as those with a 1x18 dimension. The update improved the robustness of scale handling, reducing runtime failures and contributing to more stable inference in production environments.