Inspiration
In the current ever-changing world of AI, so much of the way we live, do business, and interact with one another is changing on the daily. However, with every new and exciting technology comes bad actors. With the rise of AI, there are many scam products on the market using fake images and fake descriptions. While some more tech-savy minds may be able to pick out these scams, as AI image generation gets even more realistic, more and more people are falling victim to these scams. This issue dispropotionately affects seniors, whom were our primary customer base with this product in mind. In our personal experience, we have seen our parents and grandparents fall victim to these scams, losing out on tens if not hundreds of dollars. Our goal with this chrome extension was to help people identify if the prpduct they are planning on purchasing is using AI generated images or content to help prevent these problems in the future.
What it does
Dual-Mode Scanning: This extension analyzes both on-page text and images for AI signatures using a fine-tuned classification model. Proactive Alerts: There are automatic alerts that trigger when you visit domains with a high history of AI-generated content. Custom Brain: Powered by a custom model trained on 100,000+ samples, with a 98% accuracy rate, hosted on Hugging Face. AI vs AI: Uses Google Gemini for image processing to identify common AI image signatures
How it works
HEYI acts as a digital filter between you and the products you are planning on buying. When you visit a website, the Chrome extension "scrapes" the text and image data and sends it to a Node.js server. This server acts as a middleman, checking a MongoDB database to see if the site has been scanned before. If it’s a new site, the server queries a custom-trained AI model on Hugging Face to calculate a probability score. If the content is flagged as AI, the extension alerts you immediately with a popup or a native notification, while saving the result to your database so that future visits are instant and "proactive."
Challenges we ran into
We ran into challenges when interfacing with MongoDB. The port MongoDB uses to communicate with Node.js was blocked on our school wifi, so we instead hosted our server using Render to fix this problem. We also ran into issues when try to interface with the model we trained. There were some permission issues we were having trouble bypassing, but we got around this by hosting our model publicly on Huggingface.
Accomplishments that we're proud of
We were able to use Kaggle Notebook and a free public dataset to train an AI model on over 100,000 human and ai generated text. This took multiple hours of training and lots of troubleshooting, but we were able to get a model with 98% percent accuracy for text detection
What we learned
We learned about MongoDB and NoSQL databases, as well as about tools like Huggingface, Gemini API, and chrome web extension development
What's next for HEYI
We are currently attempting to get this actually published on the chrome extension store so it can be publicly available!
Log in or sign up for Devpost to join the conversation.