OpticGuard

• 

  End AI Cheating

Inspiration

The inspiration for this project came from two things: the difficulty computer science majors in our generation face in finding jobs, and the rampant cheating in the interview process. Hidden AI cheating tools like Cluely reward deception and undermine the integrity of our field. We wanted to help stop this, support the computer science community, and stand up for our fellow honest hackers. :)

What it does

OpticGuard is the world’s first anti-AI interview integrity platform designed to stop hidden cheating tools during interviews. By combining eye-tracking technology with audio signal processing, Optic Guard detects signs that a candidate may be using concealed AI assistance in real time. This gives employers a powerful new layer of protection for remote hiring, helping restore integrity to the interview process and enabling fairer, more confident hiring decisions.

How we built it

We first came up with the idea of using eye-tracking technology to detect whether a person was reading from a screen. Since hidden AI cheating tools are undetectable through screen sharing, we realized the only way to identify them was through external signals outside the screen itself. That led us an approach where we would scan certain varieties of retina movements with the chrome mediapipe plugin, which we later strengthened with audio analysis to gather additional data points that may indicate cheating. We also built a polished landing page to improve the product’s visibility and presentation. Overall, our app is not a catch-all solution for cheating, but the confidence score it generates can serve as a useful baseline for interviewers when evaluating interview integrity. Note: More implementation information in our github and readme: https://github.com/nickd16/optic-guard

Challenges we ran into

The hardest part was tuning the algorithms to work reliably. We wanted to keep the tool extremely lightweight and browser-efficient, so we chose simple, efficient methods over heavier approaches, since simplicity can often be more powerful. We spent a great deal of time testing different thresholds and parameter values before settling on our own algorithm for robustly detecting when someone may be reading from a screen based on eye tracking. The audio feature was especially challenging. While a neural network would normally be a natural choice, we wanted to stay aligned with the anti-AI spirit of the track and keep the product lightweight. Instead, we used classic signal processing techniques, including Fourier transform analysis and thresholding, which performed well enough for our scope given the time constraints.

Accomplishments that we're proud of

We are very proud that we were able to make this work in a way that is easy to host, robust to noise, and effective for its intended goal. We see it as an important first step toward preventing AI cheating from continuing to affect our field.

What we learned

We learned that audio signal processing, especially determining whether someone is speaking naturally or reading from an AI cheating tool, is extremely difficult without a neural network. But in the spirit of the anti-AI track and within our time constraints, we treated that challenge as an opportunity to push classic signal processing techniques as far as possible without relying on deep learning.

What's next for Optic Guard

What’s next for OpticGuard is building name recognition and eventually expanding into platforms like Google Chrome as an extension and Zoom as a plugin. We chose to launch it as a website because their APIs currently restrict the kind of eye tracking our approach requires. Our app is fully open source, and all data is deleted instantly and never stored. As we build trust with companies and users, we hope to offer OpticGuard as a practical tool for real interviewers across Chrome, Zoom, and other platforms commonly used in the hiring process.

Built With

• chromeapi
• css
• git
• html
• javascript
• mediapipe