Reverse Researcher

Start with a conclusion, not a question. Reverse Researcher uses Perplexity’s Sonar API to find supporting and opposing evidence — fast, cited, and balanced.

Comment 1

Inspiration

📖 About the Project

🔥 Inspiration

We’re surrounded by more content than ever—but that doesn’t mean we’re better informed. In classrooms, boardrooms, and online discussions, we often start with a conclusion and scramble for evidence. That got me thinking:
What if research could work in reverse?
Instead of asking a question, what if we asked AI to evaluate a claim—critically, fairly, and with citations?

Enter Reverse Researcher: an AI tool that flips the research process. Fueled by Perplexity’s Sonar API, it empowers users to see both sides of any conclusion—backed by real-time sources.

🧠 What I Learned

This project was my crash course in:

  • Real-time prompt engineering for factual accuracy and citation clarity
  • Handling user ambiguity in natural language inputs
  • Designing a research UI that balances simplicity and transparency
  • Fine-tuning API interactions with Perplexity Sonar for follow-ups and reasoning

I also learned how important it is to guide models with strong system prompts and trust-building features like citation verification and chain-of-thought reasoning.

🛠️ How I Built It

Tech Stack

  • Frontend: React + Vite
  • Styling: TailwindCSS
  • AI Backend: Perplexity’s Sonar API (sonar-small-chat)

Core Features

  • Takes any conclusion as input
  • Returns supporting and opposing evidence, using Sonar's real-time web search and citation capabilities
  • Users can select their research angle: Support Only, Oppose Only, or Balanced
  • Evidence includes citations and source credibility explanations
  • Supports follow-up questions for deeper exploration
  • Allows export to PDF

🧗 Challenges Faced

  • Model Access: I initially used an unsupported model (sonar-research-pro) and hit API errors until I aligned with sonar-small-chat
  • Citations & Reliability: Getting citations to display cleanly while preserving chain-of-thought clarity required prompt tuning and layout iterations
  • Design Simplicity: Creating a clean UX while keeping advanced features like follow-up, export, and multiple views was a balancing act
  • Real-time Debugging: Handling API rate limits, 400s, and malformed responses gracefully was critical to build user trust

🚧 What’s Next

Reverse Researcher has strong potential for growth and impact. Some of the features I’d love to explore next:

  • Citation Filtering: Let users choose academic-only, news-only, or mixed source types
  • Bias Detection: Flag possible source or phrasing bias in support/opposition arguments
  • Browser Extension: Enable one-click research on any sentence from a web article or tweet
  • Team Mode: Support multi-user sessions for classroom debate prep or peer reviews
  • Custom Agents: Let users tune tone or depth: academic, journalistic, legal, etc.

💡 Final Thoughts

Reverse Researcher isn't just a hackathon project—it's a small step toward AI-enhanced critical thinking. I’m proud to contribute a tool that doesn’t just give answers, but reveals complexity.

If you’re a student, teacher, debater, or just curious about the world—you deserve more than one side of the story. Reverse Researcher gives you both.

Built With

Share this project:

Updates