Vibe Checkr

Inspiration

The internet is a negative place and sometimes you just need a break. Wouldn't it be great if there were some way to automatically filter out all the negativity?

What it does

VibeCheckr is a Chrome extension that uses sentiment analysis to blur out all negativity you may encounter on the internet. There is a sensitivity slider as well so you can decide exactly how much negativity you can tolerate at any given time.

How we built it

The Chrome extension separates a webpage into individual sentences and passes them to a Flask server, which uses a Hugging Face sentiment analysis pipeline to determine how positive or negative each sentence is. Using the threshold specified by the user, the extension changes the style of negative sentences to blur them out. We made the front end with Bootstrap.

Challenges we ran into

Sending messages between the extension popup and content scripts
Making the button and slider states persist when the popup closes or when the current tab is changed
Sending requests from the extension to the Flask server
Incorporating ML to determine if text is positive or negative
Using Git and Github in a collaborative setting for the first time and overcoming many of its quirks

Accomplishments that we're proud of

This was the first hackathon that any of us had done, so we're really proud to have made a working product. We also take pride in persisting through the challenges and making a usable end product.

What we learned

This project was a great lesson in combining different libraries and languages to make one cohesive project. While some of us had used Git and GitHub before, we had never used it in collaboration with others, so we came out of this project having a newfound appreciation for team synergy. We also learned more about fetch requests and the importance of JSON when communicating between extension and server.

What's next for Vibe Checkr

First, we want to deploy the Flask server and publish the extension to the Chrome store so anyone can use VibeCheckr! We also would like to fine-tune the sentiment analysis model to make it faster and more accurate. While we can effectively block out negative words, what about images? We would like to use another machine learning pipeline to detect unsavory content in images and blur them out as well.