🎩 Trust Me Bro

A lie-detecting hat incorporated into a high-stakes game of deceit and wagering.

Inspiration

We drew inspiration from classic bluffing games like poker, spy thrillers, and truth-or-dare, then supercharged them with modern AI and biometrics.

For HackBrown 2026, we combined MLH’s Gemini API, Presage-style biometric analysis, and Capital One’s financial track into something playful yet technical: a wearable Lie-Detecting Hat that turns deception into a high-stakes, head-to-head game.

What It Does

Trust Me Bro is a 2-player bluffing game.

One player wears the AI-powered hat (with an embedded camera).
The opponent answers questions, sometimes telling the truth, sometimes lying.
The hat analyzes the opponent in real time and outputs a Live Lie Probability score.

Signals Analyzed

The system fuses multiple deception signals:

🎭 Facial micro-expressions & body language
❤️ Simulated heart rate (BPM) & breathing / stress
PresageTech-inspired physiological extraction
🗣️ Speech patterns & linguistic cues
Powered by Google Gemini 1.5 Pro

Gameplay

The hat wearer wagers chips on either:
- Call Lie!
- Trust ’Em
Correct calls win the opponent’s wager.
Incorrect calls cost chips.
The hat delivers snarky voice commentary throughout the game.
First player to bankrupt the opponent (or survive the most rounds) wins.

How We Built It

We built a modular full-stack system:

🧢 Hardware Layer (Python + OpenCV)

Webcam video capture
Simulated Presage-style physiology extraction:
- Heart rate & breathing from facial and chest motion
- HRV and stress proxies
Facial micro-expression detection
Speech-to-text processing
Google Gemini API for deception cues:
- Hesitations
- Contradictions
- Vague or evasive language
Streams live JSON telemetry via REST / WebSockets

🧠 Backend Layer (Node.js + Express + WebSockets)

Game orchestration (rounds, questions, wagers, chip pots)
Multi-modal data fusion into a final Lie Probability
Wager resolution and pot transfers
Win condition detection
Real-time state synchronization to all clients

🖥️ Frontend Layer (React)

Live video feed
Biometric dashboards (HR, breathing, stress)
Lie Probability gauge
Chip pots and wager UI
Dramatic reveal animations
Optional spectator mode

Challenges We Faced

Extracting believable biometric signals using only a consumer webcam
Maintaining low-latency Gemini API calls during live gameplay
Balancing the Lie Probability:
- Too accurate = no fun
- Too random = feels unfair
Combining signals from Gemini and Presage
Building the hat