Cognit

The world's first "Proof-of-Learning" protocol. We use biometric eye-tracking to prove you actually read the material, learned it, and we mint your certificate on the Kairo blockchain.

Comment

Inspiration

In a world where AI takes tests instead of people, how do you verify knowledge?

To answer this question, we built Cognit—the world’s first "Proof-of-Learning" protocol. We shift the focus from tracking simple inputs (clicks) to actual cognition using biometric hardware-backed verification.

What it does

Cognit is a "Proof-of-Focus" validation layer.

  1. Eye Tracking: We use a Tobii PC Eye 5 to track exactly where you are looking on the screen.
  2. Skim Detection: If you scroll past text without reading (missing horizontal saccades), the system locks the "Next" button. You can't proceed until you actually read.
  3. On-Chain Verification: Once you finish, we bundle your biometric data into an "Attention Score." If you pass, we mint a Soulbound Token on Kairo, proving you did the work.

How we built it

  • Hardware: We run a local Python script that interfaces with the Tobii SDK. It opens a WebSocket and streams raw (x, y) gaze coordinates to our frontend every 100ms.
  • The Wood Wide AI model: We don't just check if you're looking at the screen. We batch the gaze data and send it to Wood Wide AI. Using Wood Wide AI, we trained a prediction model using 60,000 data points and use the model to determine if the user was actively parsing/reading the content. If not, their certificate is not minted on the blockchain.
  • The Certificate (Kairo): We deployed a smart contract on Kairo that validates the Wood Wide score. It only allows minting if the "Focus Threshold" is met.
  • Frontend: Built with Next.js. We used D3.js to render the realtime heatmap overlay on top of the PDF viewer.

Challenges we ran into

  • Coordinate Hell: Mapping raw eye-tracker coordinates (screen space) to the PDF text elements (DOM space) was a nightmare. We had to write a custom calibration offset so the heatmap didn't drift when the user scrolled.
  • PCEye: The device is data locked and we had to reverse engineer the executable to intercept the data stream from the headset.

Accomplishments that we're proud of

  • The eye tracking is really good!
  • We were able to use Kairo to successfully stamp and mint contracts onto the blockchain!

What's next for Cognit

  • Accessibility: Integrating ElevenLabs to read the text out loud if the system detects your eyes are getting tired.
  • Integration: Building a plugin for platforms like Coursera or OSHA safety portals.

Built With

Share this project:

Updates