I don't disagree with the premise in your README - the first four bullet points, at least - but how is this an "invisible" AI? Doesn't this still have the exact same problem of the interviewee quiet obviously reading off of what the AI tells them, with the tell-tale eye motion, pause to get the answer from an LLM, etc?