Perspective

• 

  Browser Extension Results

Inspiration

Every news article tells a story, but rarely the whole story.

Most tools help readers consume information faster through summaries, highlights, or chat. However, a summary still inherits the framing of the original article. To understand a complex issue, readers often need to open several sources and truly understand the different perspectives that exist, many times outside of the article.

We built Perspective to address that gap.

The core idea is simple: instead of asking, “What does this article say?”, Perspective helps readers ask, “Who sees this issue differently, why do they believe what they believe, and what should I investigate next?”

What it does

Perspective turns any news article into an interactive map of viewpoints.

It surfaces: -key stakeholders -their reasoning and priorities -missing voices -questions worth exploring next

Users can then open a Research Assistant to dig deeper into any perspective.

Perspective doesn't tell users what to think—it helps them discover viewpoints they may never have considered.

How we built it

OpenAI tools everywhere. (Seriously.)

Our workflow looked like this:

Initial idea → Used ChatGPT 5.6 to compare concepts and strengthen the product direction

Architecture planning → Mapped article extraction, structured AI analysis, state management, and research flows

Feature breakdown → ChatGPT turned the full project into smaller milestone prompts for Codex

Implementation → Codex built each focused milestone while ChatGPT retained the broader product context → For inference in the extension, we used GPT-5.6 Luna, whose low latency and lower token cost made it a great fit for an interactive browser experience. We did allow this to be an editable field in settings to try out different models.

Mock generation → After getting an ugly working prototype, we used the Figma MCP integration so ChatGPT 5.6 could create and revise user friendly interface mocks directly in Figma

Final UX and testing → Refined the experience through multiple design iterations, while Codex generated automated tests to prevent regressions

This workflow let ChatGPT think about and design the product while Codex focused on writing it.

Challenges we ran into

The biggest challenge was making Perspective feel different from an AI summarizer. We repeatedly refined our prompts and UX so the focus stayed on stakeholder motivations, missing voices, and competing perspectives instead of simply restating the article. We have also seen a few news sites where this tool does not work due to how the text is presented on the screen, along with ads and other media, which have required better content extraction - which is still not perfect.

We also learned a lot about building with AI development tools:

• As new features were implemented, regressions occasionally appeared. We updated our workflow so Codex explicitly ran the automated test suite after every completed milestone before considering a task finished.
• Figma MCP sometimes generated designs in unexpected locations, so we learned to provide the exact frame URL and more specific instructions to consistently place mockups where we expected.
• While integrating the OpenAI Responses API, we encountered intermittent failures and edge cases that required creative retry and recovery mechanisms to make the extension feel reliable.
• Finally, balancing depth with speed was an ongoing challenge—we wanted the AI to provide meaningful perspectives while keeping responses concise enough to not crowd the browser sidebar. We also used structured outputs and runtime validation so the UI receives predictable, typed AI responses rather than parsing free-form text.

Accomplishments that we're proud of

We are proud that Perspective became a cohesive research experience rather than a collection of AI features.

It can:

• analyze an article in one click
• surface perspectives the reader may not have considered
• highlight voices missing from the story
• launch perspective-specific research
• preserve conversations by browser tab
• return structured, validated AI output
• remain stable through a growing automated test suite

What's next

Next, we want to add:

• move to a hosted backend so users can use Perspective without supplying their own API key.
• perfect text extraction
• live web research with citations
• side-by-side article comparisons
• perspective changes over time
• saved and shareable research sessions
• support for reports, essays, and policy documents

The long-term goal is to make Perspective a layer of the browser that helps people understand more than the single viewpoint in front of them.

Built With

• figma
• openai
• react
• tailwind
• typescript
• vite