Inspiration

Consumer fraud is becoming increasingly sophisticated, and most people only learn how scams work after they've already been targeted. We wanted to create a safe environment where users could experience realistic scam scenarios without any real risk. Our goal was to turn scam awareness into an interactive training experience instead of another article or checklist.

What We Built

We built Foolproof, an AI powered scam training platform that simulates realistic conversations with scammers. Rather than relying on scripted chatbots, our system uses multiple AI engines that work together to generate believable scam conversations, evaluate user responses, and provide personalized feedback.

The platform includes:

AI-generated scam simulations
Real-time conversation analysis
Personalized feedback after every scenario
Multiple training scenarios covering different types of consumer fraud
A challenge mode ("The Gauntlet") where users can practice against increasingly difficult scams

How We Built It

We designed the system around three specialized AI components:

Conversation Engine that generates realistic scam conversations.
Independent AI Judge that evaluates the user's decisions without bias.
Deterministic Logic Engine that tracks objectives, scoring, and progression.

Separating these responsibilities prevents the conversation model from grading itself and produces more consistent, trustworthy evaluations.

Challenges

One of our biggest challenges was making the scam conversations feel realistic while ensuring the AI remained safe and educational. Another challenge was designing an evaluation system that fairly judged user decisions instead of simply comparing them to predefined responses. Building multiple AI systems that worked together consistently also required careful architecture and prompt engineering.

What We Learned

Throughout this project, we learned how important multi-agent AI systems are for building reliable applications. We also gained experience designing AI evaluation pipelines, creating engaging educational experiences, and balancing realism with user safety. Most importantly, we learned that interactive practice can teach people to recognize scams far more effectively than passive educational content.

What's Next

We plan to expand Foolproof with more scam categories, adaptive difficulty based on user performance, voice call simulations, multiplayer training exercises, and analytics that help organizations improve employee fraud awareness over time.