You don’t get better at hard conversations by reading about them. Here you practice one with a customer named Jordan, who is having a bad week and does not care that you are new. A rubric scores what you actually said, and you can rerun it as many times as you like.
This demo is running live: Jordan is played by an LLM working from a persona brief with guardrails, and your transcript is scored against the same rubric you see on the right. Every rep is different, and every score is explainable.
{{ scriptedNote }}
I designed for private reps because role-playing in front of peers is performance, not practice. Removing the audience lets people try the sentence they would be embarrassed to fumble in a workshop.
I wrote the rubric to score only what a transcript can prove. Did you name the feeling, say what happens next, and avoid promising what the system cannot keep. "Sounded empathetic" appears nowhere in it.
Human role-play partners drift. Some go easy, some go brutal. I wrote a persona brief that holds difficulty constant across a thousand learners, which makes scores comparable and progress measurable for the first time.
People do not practice honestly while being watched by peers who will remember. A private AI partner removes the social risk, which is what makes real rehearsal possible.
Every rubric line is an observable action a transcript can prove. Vague criteria produce vague feedback, so "sounded empathetic" was banned from the start.
Unlimited reps with consistent difficulty and immediate scoring is the closest a conversation skill gets to a practice range. The persona brief keeps the opponent identical for every learner.