PERSUADE.AI

• 

  Landing Page

• 

  Profile

• 

  Stats

• 

  Achivements

• 

  Game Console

• 

  Leaderboard

Inspiration

The inspiration for PERSUADE.AI stems from the realization that while technical skills have rigorous training platforms, "soft" skills—specifically negotiation and social engineering—remain abstract and difficult to practice. We were inspired by the concept of "Aura" in modern digital culture—the intangible presence someone brings to a room. We wanted to quantify that presence using AI, creating a high-stakes "combat simulator" for communication that feels more like a tactical RPG than a corporate training video.

What it does

PERSUADE.AI is a multi-level persuasion simulator where users navigate complex social scenarios tailored to their profession.

• Simulated Stakes: Users must lower an AI agent's Resistance Meter before their own Stress Meter hits 100%.
• Dynamic Reputation: Your Aura Score carries over between levels; a high score makes future AI agents respect you, while a low score makes them dismissive.
• Psychological Feedback: After every session, a "Judge" AI provides a breakdown of your tactics, measuring your use of Logic, Empathy, Authority, and Scarcity.

How we built it

We utilized a multi-agent LLM architecture to create a realistic feedback loop:

• The Frontend: Designed with a Neo-Brutalist/Cyberpunk aesthetic using high-contrast dark modes to simulate a "hacker" environment.
• The Aura Engine: A backend logic layer that calculates performance based on efficiency. The gain $(\Delta A)$ is calculated as: $$\Delta A = \text{Base} + \left( \frac{100 - R_{final}}{\text{Turns Taken}} \right) \times \text{Aura Multiplier}$$
• The Dual-Agent System: One model acts as the Interrogator (using domain-specific jargon like "Technical Debt" or "Q4 Quota"), while a second hidden model acts as the Judge, monitoring the transcript to adjust the Stress and Resistance meters in real-time.

Challenges we ran into

One of the primary hurdles was Persona Consistency. Early iterations of the "Tech" AI would easily drift into generalities. We solved this by implementing "Knowledge Silos" that force the agent to stay within industry-specific constraints.

Another challenge was the Stress Meter calibration. We had to ensure that the AI didn't become "too hostile," which would lead to a "loss loop" where users couldn't recover. We implemented a "Tactical Recovery" mechanic where empathetic statements slightly lower the Stress Meter, rewarding balanced communication.

Accomplishments that we're proud of

• Real-time Meter Integration: Successfully syncing the LLM's psychological analysis with visual progress bars without significant latency.
• The "Aura" Atmosphere: Creating a UI that feels like a high-end terminal, making the act of "persuading a boss" feel as intense as a boss fight in a video game.
• The Judge Agent: The accuracy of the post-game psychological report in identifying a user's "communication crutches" (e.g., over-reliance on logic).

What we learned

We learned that AI is a perfect mirror for human bias. During testing, we noticed that "Tech" users almost never used the Empathy pillar, while "Sales" users struggled with Authority. This confirmed our hypothesis that people don't just need a place to talk—they need a place to fail safely and understand why they failed. We also mastered the art of "Prompt Orchestration"—managing two different AI personalities to interact with a single user simultaneously.

What's next for PERSUADE.AI

• Voice Integration: Using high-fidelity voice models to simulate verbal cues, where "stuttering" or long pauses could increase the Stress Meter.
• Multi-Agent Negotiations: Scenarios with three or more AI agents with conflicting interests (e.g., a Board of Directors meeting).
• Custom Scenario Builder: Allowing companies to upload their own "Battle Cards" or technical documentation to create bespoke training levels for their teams.

Built With

• ai
• medo
• openai
• react
• supabase