Bias Buster

• 

  Website

• 

  Code Snippet

• 

  Bias Buster

🧠 Inspiration

We're constantly bombarded by articles that look legit but subtly push agendas, twist facts, or emotionally manipulate us.
Existing tools like Grammarly or fact-checkers focus on grammar or isolated claims — but they don’t highlight how content misleads.
We built Bias Buster to bridge that gap: a tool that not only analyzes content but teaches readers to recognize manipulative techniques themselves.

🔍 What it does

Bias Buster:

• Accepts any news URL or pasted article text
• Detects tone, theme, and overall sentiment quality (e.g., “Read Freely”, “Major Issues”)
• Flags misinformation patterns like:
  • Loaded Language, Ad Hominem, Generalization, Appeal to Emotion, Red Herring, etc.
• Highlights exact phrases in the article and explains:
  • What the issue is
    
  • Why it’s misleading
    
  • Severity level: Minor, Moderate, Major
    
  • Confidence score of detection
• Suggests trusted sources users can check for counter-perspectives
• Includes Compare Mode to analyze two articles side-by-side

Unlike most tools, Bias Buster doesn’t just check facts — it checks the manipulation strategies behind how those facts are presented.

🛠 How we built it

• Frontend: HTML5, Tailwind CSS, and vanilla JavaScript
  
• Backend: Flask (Python)
  
• AI Core: Meta’s Llama 4 Maverick via Groq API for blazing-fast inference
  
• Scraping: requests + BeautifulSoup, with header spoofing
  
• Logic: Custom prompt engineering, regex-based JSON cleanup, and token-based LLM formatting
  

🧱 Challenges we ran into

• Sites like Indian Express block scraping from cloud platforms — had to spoof headers. Note: this worked for running locally, in the try-out link, articles from yahoo news and BBC work best.
• LLMs sometimes return malformed JSONs — Had to write cleanup regex functions to sanitize it
  
• Differentiating between strong opinion and misleading rhetoric took lots of prompt engineering
  
• Building a UI that is both insightful and intuitive — many tools feel overwhelming

🏆 Accomplishments we're proud of

• Real-time bias detection with in-text highlighting and tooltip explanations
  
• Compare mode to assess variance between two news sources
  
• Confidence scoring + severity levels for every detected pattern
  
• Generates output in seconds thanks to Groq’s inference speed
  
• Keeps explanations educational, not just functional — it’s about empowering the reader

📚 What we learned

• The most dangerous misinformation isn’t always false — it’s just framed misleadingly
  
• Users trust tools more when they explain decisions, not just output results
  
• Scraping the open web is a nightmare — fallback strategies are a must
  
• You can make AI fast, explainable, and accessible — all in under 1MB

🚀 What's next for Bias Buster

• 🧩 Chrome Extension for 1-click article analysis while browsing
  
• 🌍 Multi-language support for global misinformation detection
  
• 🧠 User feedback loop to improve detection quality
  
• 📷 Meme + visual misinformation detection
  
• 📥 PDF & newsletter analysis (email scanning too)
  
• 📱 Mobile UI + Android app wrapper

⚔️ How it's better than current tools

Stay sharp. Read smart. Bust bias.

Built With

• flask
• groqapi
• html5
• javascript
• llm
• python