I'm building a classifier for incoming nostr DMs to filter out and handle legitimate support requests...