Prism

Inspiration

As members of marginalized communities, we experience assumptions and misrepresentation daily. People form opinions about us before knowing our stories, judging us based on surface-level narratives rather than taking the time to understand our perspectives. This personal experience with prejudice opened our eyes to a broader problem: in today's digital age, everyone lives in information bubbles where algorithms feed us content that reinforces existing beliefs.

We realized that the same snap judgments people make about us are made about complex issues every day, shaped by one-sided news coverage, social media echo chambers, and confirmation bias. We created Prism to give people the tools to break free from these bubbles, just as we wish others would take a moment to see beyond their assumptions about us.

What it does

Prism is a Chrome extension that transforms how you consume online information. With a single click, it:

Analyzes any webpage in real-time using AI
Reveals multiple perspectives across political, economic, social, and cultural dimensions
Detects hidden biases by identifying loaded language, emotional framing, and omitted context
Generates critical thinking prompts tailored to the content to encourage active engagement
Preserves your privacy with zero tracking, no data collection, and no permanent storage

Instead of passively accepting a single narrative, Prism helps you see the full spectrum of viewpoints, encouraging informed understanding rather than blind acceptance of mass media messaging.

How we built it

Prism features a modular architecture split between three main components:

Frontend (Chrome Extension)

Built with TypeScript and React for a responsive, type-safe interface
Custom DOM traversal algorithms intelligently extract main content while filtering ads, navigation, and boilerplate
Vite for fast development builds and hot module replacement
Content scripts inject seamlessly into webpages for message-based extraction

Backend (Analysis Engine)

Python FastAPI provides a lightweight, high-performance REST API
Gemini API integration powers real-time perspective analysis
Custom prompt engineering and response validation ensure structured, accurate insights
Text sanitization and truncation optimize LLM processing

Shared Schema

JSON contract defines request/response structure
Enables frontend-backend compatibility across experimental branches
Supports rapid iteration without breaking changes

We designed the system for experimentation, with four parallel idea branches exploring different approaches: perspective matrix analysis, cognitive bias scoring, source comparison, and reflection mode.

Challenges we ran into

Content Extraction Accuracy: News sites, blogs, social media platforms, and documentation pages all structure their HTML differently. We had to build sophisticated heuristics to identify main content while filtering out navigation menus, ads, comment sections, and other noise.

Real-time Performance: Optimizing LLM responses while maintaining quality required extensive experimentation with prompt design, response streaming, and caching strategies.

Bias Detection Complexity: Loaded words, emotional framing, and omitted context often operate at nuanced levels that required sophisticated pattern recognition. We iterated continuously on our detection algorithms to catch increasingly subtle manipulation techniques.

Accomplishments that we're proud of

Content extraction accuracy across many websites
Complete perspective analysis
Zero data retention architecture that truly respects user privacy
Modular design enabling four parallel experimental branches

Most importantly, we're proud that Prism isn't just a technical achievement, it's a bridge toward a more informed, empathetic society.

What we learned

Building Prism taught us the technical complexity of real-time AI integration at scale. We learned how to optimize LLM prompts for both speed and quality, design modular architectures that enable rapid experimentation, and build privacy-first systems that respect users.

But the most valuable lesson was about technology's power to amplify human values. Prism transforms our personal frustration with being misunderstood into a bridge that helps everyone seek understanding before judgment.

We also learned that bias detection is as much an art as a science, requiring constant refinement and a deep understanding of how language shapes perception.