TruthShell

A Perplexity Sonar-powered fact-checker that instantly verifies claims in text, recordings, and images. It overlays credibility scores with sources in URL, empowering users with real-time truth.

Comment

Inspiration

AI already does more than most people imagine, yet many apps feel bloated. I set out to prove that a focused, multimodal tool can be both powerful and effortless. Thanks to the Perplexity Sonar API, TruthShell supports text, voice and image inputs: users can type, dictate or photograph a claim, and the app identifies it and judges its accuracy in seconds, much like SoundHound recognizes a song. It’s a pocket-sized oracle that separates fact from noise almost instantly. Beyond general fact-checking, TruthShell excels at rapidly verifying health-related claims—helping users sort reliable medical advice from misinformation, boost their overall knowledge, and maintain healthier habits.

What it does

TruthShell lets users verify any claim in three simple steps:

  1. Capture – type text, record speech, or photograph text.
  2. Check – the app sends the claim to a FastAPI backend, which calls the Sonar API.
  3. Learn – results include
    • Truth score (0-100)
    • Rationale explaining the score
    • Clickable sources that support the verdict

Each result is saved locally so it is always available offline.

How we built it

  • Backend: FastAPI handles requests, coordinates Sonar calls, and returns structured answers. Deployment is on Vercel for rapid iteration.
  • Mobile client: Native Android in Kotlin using CameraX, MediaRecorder, and ML Kit text recognition.
  • AI stack: Perplexity Sonar for fact‑checking, with smaller OpenAI models for audio transcription and image text extraction.

Challenges we ran into

  • Solo development required learning Android from scratch while building the backend.
  • Media handling meant selecting codecs, converting images to base64, and keeping payloads small enough for mobile networks.
  • UX design demanded a single‑screen flow that is intuitive for text, voice, and image input.

Accomplishments that we're proud of

  • Delivered a complete AI‑powered mobile app within the hackathon window, working alone.
  • Chained multiple LLM calls to blend speech, image, and text processing into a seamless response.
  • Deployed a stable backend with zero downtime during testing.

What we learned

Modern AI products shine when several specialized models work together. Careful prompt design, caching, and quick failover are crucial for a smooth user experience.

What's next for TruthShell

  • Gather real user feedback to refine the capture flow and cut latency.
  • Add shareable reports so users can post verified claims directly to social media.
  • Introduce multilingual support to fight misinformation worldwide.

TruthShell aims to be the SoundHound of fact‑checking: fast, reliable, and always at hand.

Built With

Share this project:

Updates