We are using some publicly available data and have labeled some data ourselves. We are not using users data for training currently.