logo

Inspiration

What began as a (half-serious) joke about enforced honesty turned into a real project. We were inspired to merge humor, accountability, and a bit of behavioral psychology to see if we could literally shock people into telling the truth. We wanted to build a functional prototype that blends AI and wearable tech in a single, memorable—and startling—experience.

How We Built It

We built a real-time, voice-powered AI fact-checking system. The process works as a high-speed data pipeline:

Listen & Transcribe: The system actively listens to the user's speech and streams it to Deepgram for high-accuracy, real-time transcription.
Analyze & Verify: The live transcript is fed to Perplexity AI, which is prompted to identify verifiable factual claims (and ignore opinions). For each claim, the AI performs a live web search to check its accuracy.
Feedback (If False): If a claim is found to be false, our backend immediately makes an API call to the user's Pavlok wearable, triggering a pre-configured electric shock.
Explanation: Perplexity will show explanation on why the statement is true/false.

The core technologies we integrated were:

Deepgram: For high-speed, streaming speech-to-text.
Perplexity AI: For the core logic of fact-extraction, real-time fact-checking, and generating humorous roasts.
Pavlok API: To deliver the physical aversive feedback.

Challenges We Faced

This was harder than it sounds. Our biggest hurdles were:

Speed: Keeping the entire "listen-to-zap" latency under three seconds.
Nuance: Training the AI to reliably distinguish a subjective opinion ("This hackathon is fun") from a verifiable, and false, factual claim ("This hackathon is in Paris").
Safety & Ethics: This was paramount. We had to build in strict user-configurable intensity limits and cooldown periods to prevent misuse and ensure the experience remained a safe, if startling, novelty.

What We Learned

This project was a crash course in system integration. We learned how to orchestrate multiple complex APIs (for voice, AI, and hardware) and manage the flow of real-time data streams between them.

Most importantly, we learned a valuable lesson about building responsibly with sensitive hardware. We were forced to think about user safety, consent, and configurable limits from the very first line of code, which is a critical perspective we'll carry into future projects.