I did it with one of the Go's PDF parsers (I think rsc has one, and ... some guy... forked it and added some features.. it's still kind of manual but worked great)