A nice case study showing one kind of thing one might wish to develop a Domain-Specific Language (DSL) for.

The authors identified a problem, namely improving the quality of automatically-produced data visualizations, and introduced a backend-agnostic DSL for describing relevant aspects of the meaning of the data to be visualized, i.e. the intent of the visualization. They then wrote a compiler that uses those descriptions to produce a concrete program for some specific rendering backend that in turn actually produces graphics:

We introduce Flint, an intermediate language for specifying charts using data semantics and simple visual encodings [and] develop a compiler that resolves and optimizes low-level chart configurations holistically from semantic intent, transpiling [sic] them to different visualization backends.