I noticed that LLMs are terrible at being decisive. To get around RLHF hedging (e.g., "it's important to consider both sides"), I built a system that forces constraints.

Instead of a chat interface, Persona AI acts as a computational decision engine. You provide the stakes and constraints of a specific problem. Under the hood, it runs the input against specific mental models (inversion, survivorship bias checks) modeled after specific operators (Musk, Thiel, etc.).

The UI enforces a strict output: a binary YES/NO verdict, a generated conviction score, and "Kill Signals" (falsifiable conditions where you must abandon the path).

I'd be interested in feedback regarding the UI approach to enforcing these AI constraints versus an open-ended chat window.