I’m confused about how they are transcribing the audio and creating the embeddings if it’s “private by design”