Inspiration
In a world powered by words, unconscious bias can creep into everything—from job descriptions to media headlines. We built BiasLens to help people see through the lens of fairness, using AI to detect, explain, and correct biased language. Our goal was to empower inclusive communication in hiring, education, journalism, and everyday writing.
What It Does
BiasLens is an AI-powered web platform that: Detects bias in input text and documents (gender, social, political, ableist, age-related, etc.) Explains the bias in human-readable form Suggests inclusive rewrites Visualizes patterns using a heatmap and pie charts on the dashboard It supports: Free text input Document uploads (PDF, .txt) Real-time confidence scoring Interactive visual analytics
How We Built It
Frontend: React.js with Material UI, Markdown rendering for formatted analysis, and pie/heatmap visualizations using MUI Charts Backend: Node.js + Express, deployed on Google Cloud Run LLM Integration: Google Gemini Pro / Gemini 1.5 Flash, with custom prompt tuning Database: MongoDB Atlas for bias datasets (GenderBias, SocialBias, Annotation) Hosting: Firebase for the React frontend Analytics Dashboard: Aggregates MongoDB insights into visual reports
Datasets Used
GenderBias — Stereotypical phrases, used for analysis and heatmaps SocialBias — Sentences labeled with intent/offensiveness scores Annotation — News/media bias annotations labeled left, right, or center
Challenges
Parsing documents like PDFs while preserving structure Creating context-aware prompts for Gemini Balancing simplicity in UI with the complexity of multi-bias detection Ensuring real-time performance with Gemini’s API responses
What We Learned
Prompt design is key when using LLMs—small changes led to big improvements in relevance and rewrite quality. Visual feedback helps users understand subtle bias much better than raw text. AI alone isn’t enough—UI/UX plays a major role in building trustworthy AI tools.
What’s Next
Support for image and audio bias detection Chrome extension to flag biased language in real-time Feedback loop to fine-tune rewrites based on user suggestions Dataset expansion with crowdsourced inputs
Log in or sign up for Devpost to join the conversation.