The prototype includes working LLM integrations — not screens of them. We wire it to a real model, with real prompts, against representative data. You see how it behaves under realistic inputs, not how it looks in a polished video.