Can someone hook that up with some grant database, then we can train models on grant-to-latex tasks!