TruthGuard

Ever wonder if an article or video is spreading misinformation? TruthGuard can parse articles/video transcripts in real time and search the web to fact-check their claims!

Comment

Inspiration

With AI-generated content and inflammatory political campaigning, misinformation is everywhere in this modern age, and it has left people scrambling to keep up in its wake. Moreover, the sheer volume of information available on the Internet makes it increasingly difficult to verify facts and discern truth from fiction. We wanted to build a tool that empowers users with AI-driven fact-checking, helping them navigate online content with confidence.

What it does

TruthGuard is a Google Chrome extension that:

  1. Scrapes articles and video transcripts in real time
  2. Analyzes content using the Perplexity API to find supporting/refuting evidence on the web
  3. Highlights potentially misleading or untrue statements in the article text
  4. Generates a concise summary report outlining the credibility of the content

How we built it

Frontend: Google Chrome Extension UI built using React, web scraping for video transcripts + article text Backend: Flask server, Perplexity API integration

Challenges we ran into

Our main challenges were related to frontend development of Truthguard due to unfamiliarity with the process of developing a Chrome Extension. We also faced difficulties generating reliable outputs from the LLM, leading to many changes in prompt/instruction specificity. It was also our first time using version control in Git while collaborating as a team, so we faced a learning curve initially with merging pull requests, branching, etc.

Accomplishments that we're proud of

  1. Successfully integrated Perplexity AI to provide evidence-backed fact-checking
  2. Successfully scraped article text and transcripts from the user's current viewed website
  3. Created a smooth and intuitive UI in a Chrome extension

What we learned

Limitations of generative AI models in following prompt instructions, Chrome extension frontend development, version control workflow using GitHub

What's next for TruthGuard

We want to investigate how we can use more advanced reasoning models to better follow prompt instructions while still preserving the evidence searching functionality provided by Perplexity and without significantly increasing token usage.

video: https://us05web.zoom.us/clips/share/BHVzMDQWSnZIb1JtRHdSeG1iOG1NYk02a0x5UQE

Built With

Share this project:

Updates