VeriData

Test the accuracy of claims on the internet and see through false advertisements.

Comment

Inspiration

The world has many pressing issues listed in the UN Sus Dev Goals: (poverty, climate change, war, etc). However, all of these issues rely on accurate and transparent media to reach public audiences and gather the necessary support to make a difference. Misinformation erodes trust, increases polarization, diverts resources, and delays progress toward achieving the UN Sustainable Development Goals. People do not have time to fact check their media in depth and do not have the resources to compare resources properly.

Check out this infographic for more on the impact of misinformation on the UN goals: https://www.un.org/sites/un2.un.org/files/information-integrity-and-sdgs-en.pdf

What it does

Veridata solves this issue by providing users with an AI tool that allows them to compare hundreds of sources within seconds, validating claims on a scale from 0-100. VeriData is the AI agent for checking the validity of all the crazy claims you see on the internet, in the newspaper, or even on a billboard off the highway. VeriData takes text or file/image input claims, and uses AI to analyze the accuracy of them.

How we built it

VeriData is built on a python backend which leverages Google Gemini API, hugging face sentiment analysis models, and fastAPI for end to end connection. The frontend uses HTML, CSS, and JS.

Challenges we ran into

  • getting the scoring model to work properly. It isn't just one Gemini prompt...
  • having enough api credits (i made 4 new google accounts for the free api credits)

Accomplishments that we're proud of

  • making a working MVP
  • meeting new people

What we learned

  • how to prompt engineer
  • how to use gemini api
  • how to not get robbed in toronto

What's next for VeriData

Expanding the commercial usability of this agent would be the logical next step. VeriData can currently validate claims made on advertisements and in marketing campaigns, but since these claims are a lot more nuanced, the model struggles to deliver at a very high accuracy. Incorporating another model trained off of pure advertising content could help to fill this gap.

Built With

Share this project:

Updates