BiasBreaker – “Smashing hidden bias in text.”

• 

  Icon for BiasBreaker

Inspiration

I was struck by how easily political speeches, news articles, and social media posts can shape opinions through subtle bias or propaganda. Words can sway perception without ever being outright lies. I wanted a way for people to spot the spin and think for themselves.

What I Built

BiasBreaker is a browser extension that helps users see through bias in text. It:

• Highlights loaded or emotionally charged words
  
• Detects subtle bias
  
• Provides a quick neutrality score
  

How It Works

• Frontend: Chrome/Brave extension
  
• Backend: FastAPI server using Hugging Face Transformers + PyTorch
  
• Approach: A hybrid of rules and machine learning to catch nuanced bias
  

Challenges

• Speed vs accuracy: Making the model fast enough for real-time highlighting
  
• Context matters: Bias often depends on subtle context
  
• Cross-browser functionality: Ensuring it works smoothly on both Chrome and Brave
  

Achievements

• Built a fully working AI-powered browser tool solo
  
• Enabled real-time bias detection with minimal lag
  
• Created a tool that empowers users to critically read political content
  

What I Learned

• How to integrate AI models with a browser extension
  
• The difficulty of detecting subtle, context-dependent bias
  
• Techniques for optimizing real-time performance in a web tool
  

What’s Next for BiasBreaker – “Smashing hidden bias in text.”

• Support more languages and text formats
  
• Refine the bias detection model with user feedback
  
• Add visualizations and history tracking of bias trends
  

Built With

• chromium
• fastapi
• html/css
• hugging-face-transformers
• javascript(extension-frontend)
• python(backend)
• pytorch