Possible improvements

return a [file] seq, making it lazy
use a stream way of reading, so the program is not limited by memory resources (and more efficient)
use an interface closer to xml-seq
hide functions that are not necessary
add end to end tests

Ideas