-
Content blurring on the social media demo
-
Content menu
-
Our API demo web app
Inspiration
Social media is well known for its flaws in moderation, negative impact on mental health, and spread of misinformation. With the rise of AI, our group wanted to see if there was a better way for users to take control of the content seen online and make falling for fake news a thing of the past.
What it does
We wanted to make an API that would help developers in making their application content safer and more reliable for their users. The API detects various content flags such as, hateful images, misinformation, harassment, and violence.
To showcase the uses for the API in various forms, we created two demo web applications. The first is a direct demo showing the JSON object returns, and allowing users/developers to see how the API works. The idea is to make it easy to understand, and easy to implement into moderation tools.
The second application was our vision of how the API could be implemented into social media. Users can select what content they don't want to see, and their feed will be filtered to blur out content with flag warnings. If they choose, they can still click on the post to un-blur it. We also incorporated misinformation warnings under posts that were detected as false information.
How we built it
We used the Open AI API and HuggingFace's inference API to call on the content detection models.
Challenges we ran into
We put a lot of time into fine-tuning the models we were using. There are many semantics as to what can be deemed as misinformation and we didn't want short comings in the models to take from the overall usefulness of the API.
Accomplishments that we're proud of
The five of us came from very different tech stacks, so we were proud of the amount of cooperation we had. Combining our skills went much smoother than expected, and we had fun while we were at it.
What we learned
As mentioned, the group had very different backgrounds in tech so we learned a lot from each other. Our main goal coming into the hackathon was learning more about machine learning and AI. Coming to the end of our 24 hours, we can definitely say we accomplished that goal.
What's next for Tweezer
We would like to make our API faster and more precise. Along the way we had many ideas for future features, including incorporating URL links for testing social media posts for content!
Log in or sign up for Devpost to join the conversation.