Real-Time AI Fact-Checking Agent

Inspiration

We saw how quickly misinformation shapes real decisions—elections, health, finance—while verification arrives days later. Scrolling culture rewards speed, not accuracy. The project was inspired by a simple gap: people should not need to leave the page to check whether a claim is true. Trust must live inside the feed.

What it does

Real-Time AI Fact-Checking Agent verifies any online post in seconds. The user clicks the extension, the agent extracts factual claims, compares them with high-reliability sources, and returns:

• Verified / Uncertain / False label
  
• Direct citations
  
• Short explanation of contradictions
  
• Timeline if the post was edited
  No summaries without evidence; every statement links to a source. ## How we built it
• Browser extension captures visible text from posts or articles.
  
• NLP pipeline splits content into atomic claims.
  
• Claims are embedded and matched against a curated vector database of trusted outlets.
  
• Reasoning layer evaluates agreement between sources.
  
• UI injects results inline within 3–5 seconds.
  

Challenges we ran into

• Social media language is noisy: sarcasm, slang, mixed languages.
  
• Defining a ≥90% trust source list without political bias.
  
• Balancing speed with depth of verification.
  
• Preventing hallucinated citations from the model.
  
• Rate limits when many users check the same viral post.

Accomplishments that we're proud of

• End-to-end verification under five seconds.
  
• Evidence-first design that never shows a verdict without sources.
  
• Works across X, TikTok captions, and news pages.
  
• Clear UX that non-technical users understood immediately. ## What we learned
• Fact-checking is primarily a data quality problem, not only AI.
  
• Transparency builds more trust than complex scoring.
  
• Users prefer explanations over binary judgments.
  
• Small friction at sharing time can reduce misinformation spread. ## What's next for Real-Time AI Fact-Checking Agent
• Add community reputation for sources.
  
• Multilingual verification for Spanish and Haitian Creole.
  
• API for newsrooms and schools.
  
• Detection of manipulated images and deepfakes.
  
• Offline cache for trending claims to scale globally.

Built With

• extension
• next.js
• openai
• pinecone
• typescript