• 

• 

  Agent Workflow

• 

  Example Output

GuardianEar 👂🛡️

“Listens like a worried grandkid.”

🧠 Inspiration

GuardianEar was born from a place of love and concern. I live far from my grandparents, and it hurts knowing I can't always be there to protect them from scammers who prey on their trust and vulnerability. Inspired by Microsoft Claimify and fueled by personal motivation, I built GuardianEar to be the ears I can't lend them every day.

🔍 What It Does

GuardianEar listens to real-time audio, transcribes it, then extracts atomic claims and concurrently verifies their truthfulness using the Perplexity Sonar API. If someone makes a misleading, false, or manipulative claim—GuardianEar flags it on the spot.
No more time for lies to grow.

🛠️ How We Built It

• 🎙️ Whisper: For real-time, accurate audio transcription.
• ⚙️ FastAPI Backend: Powers our claim extraction with Babelscape/t5-base-summarization-claim-extractor.
• 🤖 LangGraph Agent + Perplexity Sonar API: Each claim is analyzed, cross-checked, and labeled as True, False, Misleading, Outdated or Unverifiable.
• 📡 Designed to run continuously and process conversations in chunks.

🚧 Challenges We Ran Into

• Sonar API struggles with concurrency, so we had to experiment with async-safe wrappers and fallback logic.
• Ensuring atomic claims were accurately extracted from natural, often chaotic live speech.
• Making the UX friendly enough for seniors, with alerts that are helpful and not alarming.

🏆 Accomplishments We're Proud Of

• Built an end-to-end real-time fact-checking pipeline from scratch in under 72 hours.
• Developed a working prototype that successfully flagged phishing scams during phone call tests.
• Designed an interface that's both cute and calming, featuring a capybara mascot that makes users smile while staying alert.

📚 What We Learned

• How to integrate live audio processing with real-time NLP systems.
• How to use LangGraph and Sonar API effectively for nuanced fact-checking.
• How to design for empathy, especially when your user is someone you deeply care about.

🚀 What's Next for GuardianEar

• 📱 Phone Call Integration: Bring GuardianEar directly into mobile OS so it can run silently during phone calls.
• 📊 Scam Pattern Memory: Use embeddings to remember and match scam patterns for faster, contextual detection.
• 🧓 Guardian Mode: Let children monitor scam attempts and get weekly reports about their loved ones' digital safety.

Because a split second can be the difference between safety and regret.
GuardianEar ensures our grandparents are never alone in spotting lies.

Built With

• fastapi
• langgraph
• sonar
• whisper