VerifAI

Workflow(Planned)

Inspiration

In this project, we observed that information shared on social media—especially numerical and scientific information—was often incorrect or misinterpreted. This inspired us to create an agentic workflow that could fact check information.

What it does

With VerifAI, we objectively and clearly verify inaccurate numerical information in a text. It creates a list of facts, convert them into questions, do a web search using those questions, decides whether retrieved information from the web is reliable and finally, compares that information with facts in the beginning to create an evaluation report.

How we built it

We developed our own AI program, mainly using Gemini and Python. We also implemented a local model using requests library. We got help from the AI tools for library usage guidelines as well as bug fixes.

Challenges we ran into

We have encountered with inconsistencies in llm answers as well as format mismatches which made parsing really hard. One of the biggest issue, however, is the lack of API requests we had which lead us to implement a local model to our program. Despite working with very limited time and only free resources, after all, we successfully completed this project.

Accomplishments that we're proud of

We have managed to create a working prototype. Thanks to VerifAI, we can minimize the impact of misinformation on the internet.

What we learned

Throughout this hackathon, we learned how to work with AI and how to adapt it to different tasks. We learned the principles behind llms and how to crete workflows that could corporate to solve complex problems.