That's impressive! Seems like we're back to ELIZA again, only with a more versatile dataset format and better NLP/search