Back to Case Studies
AI AnalysisMedia Tech

Bias Lens

AI-Powered Media Literacy Platform

A news aggregation app that uses GPT-5.2 to detect political bias, identify loaded language, and help users consume media more consciously—without being preachy.

4 weeks
Development
20+
Sources
96%
Accuracy
Real-time
Analysis
Bias Detection
Left ← Center → Right

The Challenge

1

Non-Judgmental

Help users understand bias without telling them what to think. Present facts, not opinions about their media choices.

2

Real-Time Scale

Aggregate and analyze news from 20+ sources across the political spectrum. Progressive loading for fast UX.

3

Actionable Insights

Go beyond "this is biased" to show exactly which phrases are loaded and why—with educational context.

News Source Coverage

We aggregate from sources across the full political spectrum, scoring each from -3 (far left) to +3 (far right).

LeftCenterRight
Left-Leaning
  • NPR
  • New York Times
  • Washington Post
  • Vox
  • MSNBC
  • HuffPost
Center
  • BBC
  • AP News
  • Reuters
  • USA Today
Right-Leaning
  • Fox News
  • Wall Street Journal
  • New York Post
  • Daily Wire
  • The Blaze

Key Features

Smart Feed

Hero articles, For You section, time-based grouping

Bias Scoring

Article-level analysis with confidence levels

Loaded Phrases

Highlights biased language with explanations

Balance Chart

Visual breakdown of your reading habits

Explore Modes

Topics, trending, sources, clusters

Learn Tab

Media literacy education and quizzes

Bookmarks

Save articles with unread badges

Offline Mode

Cached articles work without internet

AI Analysis Pipeline

1. Article Scoring

When a user taps "Analyze Article," we fetch the full content and send it to GPT-5.2 with a structured prompt. The response includes an article score (-3 to +3), confidence level, and human-readable explanation.

{ articleScore: -1, confidence: "high", explanation: "Article uses emotionally charged language..." }

2. Loaded Phrase Detection

GPT-5.2 identifies specific phrases that carry implicit bias—emotional language, partisan framing, or misleading characterizations—with explanations for each.

loadedPhrases: [{ phrase: "radical agenda", biasType: "partisan", explanation: "..." }]

3. Topic Clustering

Articles covering the same story from different sources are grouped together, allowing users to see how the same event is framed across the political spectrum.

topicClusters: { "Election 2024": [NYT article, Fox article, Reuters article] }

Technical Architecture

Frontend
  • Swift 5.9
  • SwiftUI
  • @Observable
  • SwiftData
Data Layer
  • RSS Parsing
  • FeedCache
  • Disk Persistence
  • NetworkMonitor
AI Services
  • GPT-5.2
  • OpenAI API
  • ArchiveService
  • NLP Analysis
Features
  • Progressive Loading
  • Offline Support
  • Haptic Feedback
  • Pull-to-Refresh

Results

96%

Bias Detection Accuracy

AI scoring aligns with established media bias ratings from AllSides and Media Bias/Fact Check.

20+

News Sources

Full spectrum coverage from NPR to Fox News, with priority loading for fast UX.

4 weeks

Development Time

From concept to production-ready app with full AI integration and educational content.

Need AI content analysis?

We build AI systems that analyze, categorize, and extract insights from text—news, documents, reviews, and more.