I implemented Brzozowski's regex derivatives to build a regex implementation back-end. That back-end is used whenever exotic constructs (negation, intersection) appear in the abstract syntax of the regex; in their absence, the implementation falls back on the NFA-graph-based back end.
Yes, this seems to build on the idea of regex derivatives. If regex derivatives can be used to transform a regular expression into a recognizer for strings, why not transform a more general grammar into a recognizer of strings.