Artificial Intelligence
USA
17 Years
AI can fake it. This student built a way to spot it.
Hello, my name is Anishka Kallepalli. I’m a high school senior, and outside of school, my interests range from engineering and AI to debate and public policy.
I’ve always enjoyed taking problems that seem complicated and trying to figure out how technology could make them more understandable or accessible.
What problem made you decide to build AI Expose?
As AI-generated images and videos became increasingly realistic, I started realizing how difficult it was to tell what was real and what wasn’t just by looking at it. I was especially interested in how quickly AI-generated media was spreading across social media, where people often see content for only a few seconds before scrolling past it.
I wanted to build something that could give people an additional signal while they were browsing instead of making them download an image, upload it somewhere else, and try to figure it out themselves. That idea eventually became ExposeAI.
Often, detectors require a full download-upload cycle for each video. By the time that finishes, most users will have moved on.
For someone hearing about AI Expose for the first time, what does it do?
ExposeAI is a Chrome extension that helps people identify potentially AI-generated images and videos while they browse the internet. Users can analyze media directly from the webpage and receive an AI-likelihood result and confidence score without having to leave the site.
What was the hardest part about building it?
The hardest part was definitely getting everything to work together. I had to learn a lot about machine learning models, browser extensions, APIs, servers, and web development while actually building the product.
One of the biggest technical challenges was figuring out how to run the AI detection model efficiently for users without requiring every user to have their own API key. We experimented with different approaches, including ONNX, before ultimately moving toward a self-hosted classification server.
There was a lot of debugging involved, and sometimes something that seemed like a tiny change would completely break the extension. But that was also probably the most valuable part of building it because I learned how to actually troubleshoot instead of just following a tutorial.
What has surprised you the most since launching?
I was surprised with how quick the Chrome Store was to accept my extension, even as a large package.
What advice would you give to budding founders?
My biggest advice would be to start before you feel ready. I definitely didn’t know everything I needed to know when I started ExposeAI, and if I had waited until I understood machine learning, web development, servers, and extensions perfectly, I probably never would have started.
Also, don’t be afraid to build something imperfectly. Your first version isn’t supposed to be your final version. Put something in front of people, see what breaks, listen to what they say, and keep improving it.
Are there any books, podcasts, or resources you would recommend?
For actually learning how to build, I’ve found documentation, GitHub, and tutorials to be some of the most useful resources. A lot of my learning has come from having a specific problem in front of me and then researching until I understood how to solve it.
One book I’d recommend is The Art of War for Women by Chin-Ning Chu. I liked how it approaches competition, leadership, and navigating difficult situations strategically. A lot of the ideas resonated with me while building ExposeAI, especially learning to be persistent while also being strategic about which problems are actually worth solving.
Where can we find you?
The extension is also available through the Chrome Web Store, and we’re currently developing iOS and Android versions.
What plans for the future?
The goal is to make ExposeAI useful beyond the desktop browser. We’re currently working toward mobile versions for iOS and Android so users can evaluate AI-generated media wherever they’re encountering it.
We’re also continuing to improve the detection system, expand the types of media and websites it can work with, and make the results more reliable and understandable.
Ultimately, I want ExposeAI to become a simple tool that helps people pause and question what they’re seeing online rather than automatically assuming that everything is real.