So this started as a weekend thing. I wanted to practice negotiation skills — not the awkward "let me read a book about it" way, but actually doing it and getting feedback on whether I'm using logic, empathy, or just vibes. Three weeks later I have a full blown app called PERSUADE.AI and I've spent way too many hours making pixel fire follow a cursor around.
Let me explain what it actually is.

What is PERSUADE.AI ?
Link
It's a persuasion and social engineering simulator. You pick a profession — Tech, Sales, Management, or General — and you get dropped into AI-driven scenarios where you have to convince an NPC of something. Could be pushing back on scope creep with a stubborn stakeholder (very cathartic if you work in tech), closing a difficult sales call, or resolving a team conflict.
There are two meters on screen. A Resistance Meter for the AI — you're trying to get that to zero. And a Stress Meter for you — don't let that hit 100 or you lose. It's basically Dark Souls for your communication skills.
When a scenario ends, you get a psychological breakdown of how you actually argued. What percentage was logic. What percentage was empathy. How much authority or scarcity you leaned on. It's weirdly humbling.

The dual-agent setup is the interesting part
There are two AI agents running under the hood. Agent A is the NPC you're talking to — the interrogator. It stays in character, uses profession-specific language, and adapts its tone based on your Aura score (more on that in a second).
Agent B is hidden. It's watching the entire conversation transcript and acting as a judge. It decides win/loss, calculates your Aura gain, and writes the psychological report. I really liked this pattern because it separates the "character performance" from the "evaluation logic" cleanly. No awkward moment where the NPC stops mid-sentence to grade you.

The Aura system is genuinely fun
Every user starts with 100 Aura. Win a scenario and you get +20 to +50 depending on performance. Lose and it's -10. But here's what I like — the AI's starting attitude changes based on your Aura.
Under 150? The NPC is skeptical and dismissive right from the first message. Over 500? It opens with some respect. It means your past performance actually bleeds into future sessions. High Aura players are playing a different game.

What surprised me
The psychological breakdown at the end is what people keep reacting to. Not the fire, not the voice input — just seeing "you used 62% logic, 11% empathy, and almost zero scarcity" after a conversation. People immediately want to run it again to try a different approach. That loop is satisfying in a way I didn't fully anticipate when I was building it.
Also the dual-agent architecture turned out to be very clean. Would use it again for any app where you need both an interactive NPC and quality evaluation of user behavior.

The stack is pretty standard web stuff. The interesting pieces are the Web Audio API synthesis, the particle systems, and prompt engineering for the dual agents. Happy to go deep on any of those if people are curious.
Try it out if you want to find out whether you actually argue with logic or just assert things confidently and hope for the best. I learned some uncomfortable things about myself.

Drop a comment if you want to see a breakdown of the agent prompt architecture or the audio synthesis code.