I use Hachoir for a lot of tasks - digital forensics (examining file tables on a filesystem, discovering slack space taken up by hidden data, etc.); for metadata extraction (for which the hachoir-metadata subproject is designed); for file carving (either with hachoir-subfile or manually); and just to geek out about file formats (did you know .doc/.ppt is secretly a FAT filesystem?).

Funny, I had no idea it could do all that. I've been using it in production for years to extract width/height/duration of media files at compile time so my loading placeholders are perfectly sized at runtime in a basic web app.