Tripwire

Dashboard
Demo
Saved Calls

Inspiration

When something feels wrong, many of us call someone we trust. For an older adult, that might be a daughter who helps with the bills, a grandson who checks in every Sunday, or a longtime friend. Those relationships are already a powerful source of protection. We wanted to bring them into the moment when someone needs help most.

We imagined Rosa receiving a frightening call: her grandson Diego is in trouble, needs money immediately, and asks her not to tell anyone. Her instinct is to help someone she loves. But the urgency and secrecy separate her from the very people who could help her check the story.

That inspired Tripwire: a bank teller that connects the family back together. When Rosa attempts an unusual payment, the teller gently asks what it is for. After hearing her story, it asks permission to contact Diego through the trusted contact already saved on her account. One conversation—“Grandma, I’m okay. I didn’t ask for money”—can change everything.

Tripwire began with a simple idea: the people who care about you should be easier to reach than the people trying to rush you.

What it does

Tripwire is an AI-powered safety teller that steps in during high-risk payments.

It detects unusual transactions, talks with the customer in real time, identifies social-engineering signals like urgency and secrecy, and calls a trusted contact to verify the situation before releasing the money.

Instead of trying to detect whether a voice is a deepfake, Tripwire verifies the real person.

How we built it

We built Tripwire with Next.js, TypeScript, Gemini Live, ElevenLabs, Tiger Data, and Node.js.

Gemini powers the real-time safety-teller conversation, ElevenLabs powers the live scam and verification calls, and Tiger Data analyzes transaction history to identify unusual payments.

Challenges we ran into

Our biggest challenge was coordinating multiple real-time AI systems while keeping the experience fast and natural.

We also had to design Tripwire to protect users without making them feel blamed, confused, or overwhelmed.

Accomplishments that we're proud of

We built a complete end-to-end fraud prevention experience that can detect a risky payment, talk with the customer, contact a trusted person, verify the situation, and stop the transaction in real time.

We are also proud that our demo uses a live AI voice clone to demonstrate the threat—and AI to stop it.

What we learned

We learned that fraud prevention does not always require detecting what is fake.

As AI-generated voices become more convincing, verifying the real person through a trusted channel can be more reliable than trying to identify a deepfake.