The Internet, the place for good and the bad. A world away from the real world. But is it a safe world?? Stats suggest that 15% of high school students (grades 9–12) were electronically bullied in the past year, About 1 in 7 kids have been sexually solicited online, kids start watching porn at the average age of just 11! Sure, there are many parental control solutions available but they're mostly generic and super easy to go through. Even schools go through an extensive counseling to tell the students about grave dangers of the internet but peer pressure usually supersedes these educational drills and kids end up getting misdirected anyways.

What it does

It safeguards children in the world of the Internet. The internet is a scary place. During teenage, there's high curiosity in children and they could fall into these pits. Using deep learning based natural language processing and computer vision techniques, this extension filters out illicit content (both text and image based) from legitimate websites while ensuring the privacy of the user. Using Sentiment analysis, text sequence to sequence models, clickbait ad detection, NSFW Detection, feedback loops, we ensure that parental controls are not just a button that says 'I'm above 18'.

How I built it

We've used an extensive bi-directional sequence to sequence models for sentiment analysis and detecting clickbait from its meta-data. To identify levels of nudity in visual content, we are leveraging convolutional neural networks. Even while detecting illicit content we are taking care of the users' privacy by making sure that the users' data is not being transported elsewhere but rather being parsed by AI models within the browser.

Challenges I ran into

Data collection and processing was a challenge, but we were able to use AWS and Google Cloud Platforms to enable it. Privacy and data collection is a very sensitive topic and hence, we decided to run all the neural networks within the browser itself. To improve the system we used feedback loops so that the experience of the users can be continually improved.

Accomplishments that I'm proud of

Getting various tools together into the chrome extension was a big challenge, but our team was quite skillful in getting the job done. Ensuring that all the AI models fit within the chrome extensions size limit was also a major task.

What I learned

Keeping the memory and CPU consumption under control while running AI Models. Running JavaScript AI Models. Making chrome extensions.

What's next for Chrome Extension

Ensuring that the extension works at a global level, hence for most language types, for this, we will be using Google's Neural Machine Translation systems.

Built With

Share this project: