We demonstrate an end-to-end differentiable pipeline for shape optimization, serving as a template for more complex scientific machine learning workflows. Our toy example uses a simple obstacle ...