Inspiration
Imagine scrolling through your feed and a shocking headline stops you: a celebrity scandal, a major world event, a groundbreaking discovery. You’re intrigued, maybe even share it, only to later discover it was completely fabricated. This isn't just a personal annoyance; misinformation and disinformation are now recognized as the #1 global risks, even surpassing “extreme weather event” in the World Economic Forum's 2025 report.
But what if we could transform our online experience? What if every piece of content – a news article, a social media post, even a product review – was transparent, showing not just the information itself, but also its verified accuracy and the collective sentiment surrounding it?
This isn't just about debunking falsehoods; it's about fostering a more informed and connected digital world. It’s about empowering individuals to navigate the vastness of the internet with confidence, knowing they have the tools to discern fact from fiction and to see the authentic reactions of others. Truely envisions a future where trust and transparency are inherent in our online interactions. We believe everyone deserves to engage with information that is reliable and to connect with the genuine opinions of their community. Join us in building a web where truth prevails and authentic voices are amplified.
What it does
Truely is a Chrome extension featuring two core functionalities: Truely-Check and Truely-Lens. Truely-Check is a fact-checking tool that analyzes an entire webpage or selected content, identifying claims and displaying their verified accuracy. Truely-Lens reveals public sentiment towards any web content through aggregated emoji reactions.
How we built it
Truely-Check employs a dual approach to fact-checking: a knowledge base of established facts and a stringent fact-checking mechanism. Complex claims necessitate evidence and expert review. Individual content pieces, even single sentences, can contain numerous claims demanding separate verification. Many prevalent online claims have already been debunked and are aggregated in Google's Fact Check Database API. Our agentic system, utilizing Langchain for parallel processing, emphasizes efficiency. Initially, an LLM identifies claims within the provided content. If no claims are detected, it responds with no claims detected and where applicable Truely-Lens shows public sentiment. If claims are present, an LLM extracts them, and each is verified against Google's Fact Check Database API. For claims absent from the database, a preliminary fact-check process is activated (currently in development and requiring expert input). This involves an LLM formulating verification queries, a CustomGoogleSearch tool searching reputable sources (Wikipedia, .org, .edu), and subsequently, an LLM evaluating the collected data to validate the claim, producing a verdict accompanied by sources. For inference we used groq and llama3-70b model.
Truely-Lens utilizes standard software development methodologies. The location of content on a webpage is recorded using the page URL and image URL. It shows the sentiment summaries with reaction counts and top comments. For selected text, basic HTML syntax is stored. This functionality is currently implemented for image content.
Challenges we ran into
The development of Truely presented several technical hurdles.
Truely-Check: Integrating multiple Large Language Models (LLMs) for specific tasks within Truely-Check proved challenging, particularly in designing effective prompts. For instance, the initial claim extraction step struggled to produce consistently formatted, individual claims suitable for programmatic parsing and subsequent processing. Early iterations using the llama-7b model often yielded outputs that were not proper claims. This issue was resolved by transitioning to the more capable llama3-70b model. Additionally, implementing the CustomGoogleSearch tool encountered rate limit errors, despite the expectation of a substantial free usage tier.
Truely-Lens: Hosting and deploying the database for the Truely-Lens feature presented a minor challenge. Furthermore, accurately positioning the sentiment statistics ribbon at the precise corner of any image on a webpage required careful implementation.
Accomplishments that we're proud of
We are incredibly proud to have delivered a functioning and deployed prototype of Truely. With the current Chrome extension, users can simply select text on any webpage, right-click, and effortlessly fact-check against already verified claims using Truely-Check. Furthermore, our Truely-Lens feature empowers users to react to any images across the internet, allowing everyone to see what's currently trending.
We also take immense pride in addressing what the World Economic Forum's 2025 report has identified as the most critical global challenge. We are confident in Truely's market research and well-defined business model. We have also identified potential stakeholders who could benefit, such as fact-checking organizations who could use it to prioritize which claims the public suspects.
What we learned
The hackathon was an invaluable learning experience, fostering team collaboration and refining diverse ideas into a cohesive vision. This intensive event provided hands-on experience in agentic AI and prompt engineering, crucial for our project. We honed initiative-taking and integration skills, while also learning market research and pitch deck creation. We gained practical insights into problem-solving under pressure and navigated development challenges with agility, experiencing the full product lifecycle from ideation to deployment. Beyond technical skills, the hackathon highlighted effective communication, time management, rapid iteration, and leveraging diverse team strengths. The collaborative environment accelerated learning and fostered a passion for innovation.
What's next for Truely
We will gather user feedback on the current Truely functionality. Subsequently, we aim to consult a fact-checking organization to understand their workflow and compare it with our planned approach. Prior to this consultation, we intend to implement our agentic fact-checking method and present initial results for their feedback. Furthermore, we plan to incorporate fact-checking for scientific claims based on published literature. For Truely-lens, we aim to extend its capabilities to analyze text content in addition to images. We also plan to integrate a notification feature to alert users about trending topics aligning with their interests. We will seek feedback from investors through pitching Truely to refine our concept and explore adding further functionalities.
Log in or sign up for Devpost to join the conversation.