L4RPCH3KR

The real-time, AI-powered wearable BS-detector for hackathons.

Comment

Inspiration

According to Urban Dictionary, a "Larper" is someone who does something way out of their capability but genuinely thinks they can do it, even though it's obvious they're nowhere near doing so. In hackathon culture, it's extremely common for people to “larp” about their technical abilities. We’ve all met the person who claims to be a kernel developer but hasn't pushed code in three years.

We asked ourselves: What if there was a system that could detect what people are saying, cross-reference it with their actual verified credentials in real-time, and assign a “larp score” based on how much they’re lying? Thus, L4RPCH3KR was born.

What it does

L4RPCH3KR is a wearable AI-powered system designed to keep hackathon attendees and just about everyone honest. It consists of three main parts:

  1. The Larp-Checking Device: A Raspberry Pi 5 worn on the chest. It uses a camera to scan the face of the person you're talking to in order to start the process, identifies them through their audio introduction, and pulls their LinkedIn/GitHub data. A microphone listens to their claims and parses them.
  2. The Real-Time Brain: As the person speaks, the system transcribes the audio and extracts claims (e.g., "I have 5 years of experience in Rust"). It compares these claims against their scraped professional data.
  3. The Feedback Loop: If a "larp" is detected it updates the web dashboard visible online and sends an alert.
  4. The Recap: After the conversation, the user gets a summary with a "Larp Score" (labels like "Mostly Honest" or "Approaching Freestyle") and a list of flagged moments. Furthermore, their score is added to the "Larper-Board" (leaderboard) so that others can be better informed of their abilities.

How we built it

We used a "contract-first" architecture to ensure our three sub-teams could work in parallel:

  • The Backend: Built with Mongo as our database. We integrated Whisper for transcription and an LLM pipeline for claim extraction and verification. We used WebSockets to stream data between the Pi, the server, the web dashboard+LarperBoard, and the phone app.
  • The Hardware: A Raspberry Pi 5 (16GB) equipped with a Logitech camera/mic. The Pi handles the video/audio capture.
  • The Mobile App: A React application that serves as the HUD. It handles onboarding.
  • The Dashboard: A React web interface for organizers to upload attendee CSVs and manage the "Linked Database" of credentials.

Challenges we ran into

  • Latency: Our "Latency Budget" was tight. To make the alerts come in real-time so that the user could immediately get accurate information, we had to offload some processing from the backend to the Pi so that the data transfers happened with simple text data rather than full files.
  • Hardware Integration: Managing stable WebSocket connections on the Pi while simultaneously processing video frames and audio buffers required careful threading and memory management.

Accomplishments that we're proud of

  • The Closed Loop: Successfully getting a verbal lie to trigger an alert on someone's chest in under 4 seconds.
  • Hardware-Software Synergy: Seamlessly pairing a Raspberry Pi to a website and seeing the data flow across the network with minimal jitter.
  • Automated Verification: Building a scraper that takes a name and automatically generates a truth-baseline from GitHub and LinkedIn profiles.
  • QR Code Sign-in: Integrating an automatic QR code scanner that reads a code on the mobile app without prompt in order to authenticate the user of the device before they start talking to others.

What we learned

  • Real-time audio processing is significantly more complex than batch processing, especially when dealing with background noise at a loud hackathon.
  • "Contract-first" development is a lifesaver for hackathons. Defining our API and WebSocket schemas on Day 1 meant we had zero integration headaches on the final day.

What's next for L4RPCH3KR

  • Social Leaderboards: Opt-in "Honesty Rankings" for the entire hackathon event.
  • Company Expansion: Making a software version that caters to companies who want to further verify potential employees before hiring in a much cheaper and efficient way.

Built With

Share this project:

Updates