YouTube Fact-Checker

Real-time fact-checking for YouTube videos using AI.

Architecture

• Chrome Extension - Frontend overlay on YouTube videos
• FastAPI Backend - Processing pipeline with async queues
• AI Services - Whisper, RunPod Deep Cogito v2 70B, ACI, OpenAI

Backend Setup

1. Install Dependencies

cd backend
pip install -r requirements.txt

2. Environment Variables

# Copy the example file
cp env.example .env

# Edit .env with your API keys:
# OPENAI_API_KEY=sk-your-openai-api-key-here
# RUNPOD_API_KEY=your-runpod-api-key-here  
# ACI_API_KEY=your-aci-api-key-here

3. Install System Dependencies

# Install FFmpeg (required for yt-dlp audio processing)
brew install ffmpeg  # macOS
# sudo apt install ffmpeg  # Ubuntu/Debian

4. Run the Server

cd backend
uvicorn main:app --reload --host 0.0.0.0 --port 8000

5. Test the API

# Health check
curl http://localhost:8000/health

# Process a video
curl "http://localhost:8000/api/process-video?video_url=https://www.youtube.com/watch?v=jNQXAC9IVRw"

API Endpoints

• GET /health - Health check
• GET /api/process-video?video_url=URL - Process YouTube video for fact-checking

Request Format

GET /api/process-video?video_url=https://youtube.com/watch?v=VIDEO_ID

Response Format

{
  "video_id": "VIDEO_ID",
  "title": "Processed Video",
  "total_claims": 1,
  "claim_responses": [
    {
      "claim": {
        "start": 0.0,
        "claim": "factual claim text"
      },
      "status": "verified|false|disputed|inconclusive",
      "written_summary": "Detailed explanation of fact-check result...",
      "evidence": [
        {
          "source_url": "https://example.com",
          "source_title": "Source Title",
          "snippet": "Evidence excerpt..."
        }
      ]
    }
  ]
}

Frontend Setup

We created interfaces.

Interface 1: Chrome extension v1 (demoed)

Once you have the backend app running, you'll need to unpack the Chrome extension

1. Go to Chrome, navigate the manu towards Extensions and then Manage Extensions
2. Switch on the Developer mode toggle
3. Click on Load unpacked and unpack the chrome-extension in the frontend project directory: Stockholm-Hackathon/frontend/public/chrome-extension
4. You'll then see YouTube Fact-Checker extension. Re-open your browser.
5. Go to Youtube, play a video, then click on the liquid-glass circle in the upper-right corner to activate fact checking

Interface 2: Chrome extension v2

Once you have the backend app running, you'll need to unpack the Chrome extension

1. Go to Chrome, navigate the manu towards Extensions and then Manage Extensions
2. Switch on the Developer mode toggle
3. Click on Load unpacked and unpack the chrome-extension in the frontend project directory: Stockholm-Hackathon/yt-fact-checker-extension
4. You'll then see YouTube Fact-Checker (Live) extension. Re-open your browser.
5. Go to Youtube, play a video, the fact-checker pop-up will be immediately available. Note that it takes a few seconds to start transcribing

Interface 3: Dashboard

Run the backend and frontend in separate terminals.

1. Start the backend API (terminal A)

cd backend
uvicorn main:app --reload --host 0.0.0.0 --port 8000

2. Start the dashboard UI (terminal B)

cd frontend
npm run dev

Then open http://localhost:3000 in your browser. Paste a YouTube URL, click Analyze, and watch claims and fact‑checks appear when processing completes.

Processing Pipeline

1. Transcription - Download audio with yt-dlp → OpenAI Whisper API → Sentences with timestamps
2. Claim Extraction - RunPod Deep Cogito v2 70B → Extract factual claims
3. Evidence Gathering - ACI + EXA_AI → Find web sources and evidence
4. Fact-Checking - OpenAI GPT-4 → Analyze evidence → Return verdict

Tech Stack

• FastAPI - Async web framework
• OpenAI Whisper - Audio transcription with timestamps
• RunPod Deep Cogito v2 70B - Advanced claim extraction
• ACI + EXA_AI - Evidence gathering and web search
• OpenAI GPT-4 - Fact-checking analysis
• yt-dlp - YouTube audio download
• Pydantic - Data validation and structured outputs

Development

The backend uses async queues for concurrent processing:

• Sentences are streamed from transcription
• Claims are extracted in real-time
• Fact-checking happens concurrently
• Results are returned as structured JSON

Perfect for hackathon development with clean separation of concerns and production-ready architecture.

$env:PATH += ';C:\ffmpeg\bin' cd .\backend\
.\venv\Scripts\activate python -m uvicorn main:app --reload --host 0.0.0.0 --port 8000