For our deep learning models, a data-first approach to ML infrastructure bridged the gap between Data Engineering and ML