ForReal 🔍
Real-time AI Fact-Checking & Media Verification for X (Twitter)
Inspiration
In an era where misinformation can spread faster than facts, discerning truth from fiction on social media has become an increasingly difficult task. The sudden rise of deepfakes and AI-generated content has only further blurred the lines of reality, leaving many users unsure of what to trust. We were inspired to build ForReal to address this issue directly, empowering users with immediate, reliable information without requiring them to leave their timeline. Our goal was to create a tool that feels like a natural extension of the browsing experience.
What it does
ForReal integrates seamlessly into the X (Twitter) interface by adding a discreet Verify button to posts. When a user questions a tweet and clicks this button, the extension springs into action, performing a multi-layered analysis in real-time. It doesn't just look for keywords; it analyzes claims and cross-references them with reputable sources using the Brave Search API, while simultaneously employing Google Gemini AI to synthesize a clear verdict, whether that be True, False, Misleading, or Unverifiable. Beyond text, it also runs media forensics to detect AI-generated imagery and analyzes the framing of the content to highlight potential political bias, ensuring users get the full picture before they share or react.
How we built it
We built ForReal using a modern, modular architecture that keeps the browser extension lightweight while offloading the heavy analysis to a robust backend. The frontend was developed using Manifest V3 for security and performance, utilizing standard JavaScript to inject our custom UI components directly into the X feed. On the server side, we utilized Python and FastAPI to create a high-performance, asynchronous REST API. The true intelligence of the system comes from chaining together three powerful services: Google Gemini acts as our reasoning engine, Brave Search provides unbiased real-time data, and the AI or Not API handles the complex computer vision forensics required to spot deepfakes.
Challenges we ran into
Our biggest technical hurdle was undoubtedly the DOM manipulation on X. Since the platform uses dynamic class names that change frequently, it was incredibly challenging to consistently inject our buttons and read tweet content without the extension breaking. We had to implement robust selectors and observers to ensure stability. Additionally, we spent a significant amount of time on prompt engineering; tuning Gemini to provide concise, accurate verdicts without hallucinating took many iterations. We also had to carefully optimize our backend to manage rate limits for our search and AI services to ensure the user experience remained snappy despite the heavy processing happening in the background.
Accomplishments that we're proud of
We are incredibly proud of how seamless the integration feels. Getting the Verify button to look and behave like a native part of the X interface, despite the technical hurdles of DOM injection, was a massive win for us. We are also proud of the backend engineering required to successfully chain text analysis, web search, and image forensics into a single, fast request. It was satisfying to optimize that pipeline so that users receive a comprehensive verdict in mere seconds, maintaining the flow of their social media browsing without interruption.
What we learned
This project taught us that while LLMs like Gemini are powerful, they require precise instruction to act as objective fact-checkers, giving us deep insights into context injection and prompt structuring. We also learned a great deal about modern browser security standards by working within the constraints of Manifest V3. Furthermore, handling the messy, unstructured data of social media posts and converting it into structured queries for our APIs gave us a new appreciation for the complexities of real-world data processing.
What's next for ForReal
Looking ahead, we plan to expand our media checks to include video analysis, which is becoming a leading source of misinformation. We also aim to bring ForReal to other platforms like Instagram, Reddit, and LinkedIn to fight misinformation wherever it appears. Finally, we plan to implement a caching system for frequent requests, allowing us to serve accurate information to users instantly while eliminating expensive backend calls.
Log in or sign up for Devpost to join the conversation.