HealthShield

Inspiration

No matter how much the current generation knows to follow verified health information, there are still individuals who like to get their information from social media channels... like Twitter. But unfortunately, such channels contain accounts which spew misinformation, pseudoscience and carry out fear mongering. We wanted to help individuals still use the familiar social media apps they know and love, while also helping them gain fact-based insights about their health and not get subjected to such stressors. Seriously, who wants to see things like "Ginger can cure COVID-19" when you're just trying to look at your friend's account?

What it does

HealthShield is an AI-powered chrome extension that detects health misinformation on Twitter in real-time. It helps users filter out misleading medical advice, fear-mongering, and pseudoscience while promoting fact-based health discussions. It leverages local machine learning models to analyze and rate the credibility of health facts. This tool is a must-have for anyone seeking reliable health information on the internet, particularly for those from marginalized communities who often face health disparities.

How we built it

We used Python for the backend, using a BERT model from the Hugging Face library which we fine tuned with some relevant health datasets to make the model as accurate as possible for our purposes. We set up an API endpoint in the backend using Python and Ngrok. We used pure JavaScript for the frontend.

Challenges we ran into

It was a challenge finding the proper datasets (and amount of data) to fine tune the LLM without the training taking way too long. We were also using Google Colab's free GPU's and it would frequently rate limit us - which required some workarounds.

Accomplishments that we're proud of

Given that this was my first time making a functional ML application using Python, we were able to get a working locally fine tuned model with an accuracy of 77%.

What we learned

How to use Python's various machine learning libraries JavaScript is not as bad as we thought :)

What's next for HealthShield

• Add functionality for multiple languages within Twitter
• Add functionality for multiple social media platforms (eg: Reddit)
• Perhaps fully train and fine tune a local LLM for more personalized insights

Built With

• ai
• api
• chrome
• javascript
• llm
• python
• r