That's not really the same approach. It's not "backward", it's more "sideways": It starts with an end-to-end compiler for a tiny language, then builds out that end-to-end compiler incrementally. The featured article describes an approach where no such end-to-end compiler is available from the beginning. It's only end-to-end after the final exercise.

They link this paper in the bottom of the article as a different approach from the one they're using.