TruthMeter

• 

  Logo

Inspiration

Although we don't have a specific inspiration, our team member Anthony came up for the idea for this project. We wanted to do something that was more than just a simple program for our first project and had never done extensions before, so we gravitated towards that.

What it does

The extension looks at the website you are currently on and employs LLMs to assign it a credibility score. The output is shown directly in the extension's sidebar, along with explanations for the score of each of the claims and the overall score.

How we built it

Our solution uses the tools provided by Google ADK to design an easy-to-use Chrome Extension that scrapes the text from a given website and sends it to multiple Looped Agents that break about the claims on the website and subsequently evaluates the credibility of said claims using the Agent Tool protocol, allowing students to easily check if a source is credible.

Challenges we ran into

There were numerous challenges we faced, most of which involved debugging. Some of our favorite issues were incorrect import structures, class variables being declared wrong, and Marty's personal favorite, getting rate limited by Google. Luckily, after implementing a timer and using code tracers, most issues were able to be solved by looking at console logs.

Also, make sure to include .env in .gitignore!

Accomplishments that we're proud of

This is the first Hackaton for all of us, and being able to submit a functional project is a reward in and of itself. We managed to not only get AI tools to work, but also created UI and made it so all of the front-end and back-end parts interacted. The most exciting break-through moments was finishing the AI agents, creating the Python pipeline, and finalizing the extension's UI.

What we learned

The most significant thing our team learned was how to deploy Google ADK protocols and tools in our own projects. We also learned how to design Agent Tools, make React code, and pipe variables in Python.

We also learned that it is important to start the debugging process early to avoid getting bottlenecked. AI tools such as Gemini were also extremely helpful for debugging on top of being the basis for our program.

What's next for TruthMeter

If we had more time, improving the efficiency of the agents to make less calls would be the first thing we would do. The UI could also be revised to be more ~exciting~.

Built With

• css
• gemini
• javascript
• python
• react