🕵️‍♂️ VidScrutiny: AI Fact-Checker

Real-Time "Community Notes" for YouTube Shorts & TikToks

Text fact-checking is a solved problem. Video is the final frontier.

VidScrutiny acts as an automated, real-time "Community Notes" overlay for short-form video content. It digests video streams in seconds, transcribes the spoken audio, cross-references claims with the live internet, and projects beautiful, broadcast-quality fact-check overlays directly onto the video exactly when the claims are spoken.

✨ The "Wow" Factor Features

• ⚡ Zero-Latency Feel: Combines Whisper-v3 and Llama-3.3-70b on specialized Groq hardware to process entire Shorts/Reels in seconds.
• 🌐 Live Web Search (RAG): AI hallucinations are neutralized. The system actively queries DuckDuckGo behind the scenes to find real, live URLs to back up or debunk every claim.
• 🇮🇳 Multilingual AI: Natively supports translating English videos into Hindi fact-checks (and vice-versa) via a single UI toggle.
• 🚀 Real-Time WebSockets: No fake loading bars. The Python backend streams its exact processing state to the frontend via WebSockets.
• 💾 Smart SQLite Caching: Analyzed a viral video once? It's cached in the database. The next user who pastes that link gets the result instantly.

---

🏗️ Architecture Pipeline

1. Ingestion (yt-dlp): Securely bypasses bot-blockers to extract raw .mp4 and .m4a streams.
2. Transcription (Whisper API): Generates a highly accurate, timestamped transcript array.
3. Contextual Search (DuckDuckGo): The LLM isolates the main topic and silently searches the internet for real-time context.
4. Analysis (Llama 3): Rates each claim as True, False, or Context and formats it into strict JSON.
5. Presentation (HTML5/Tailwind): A custom video player triggers sleek, Fitts's Law-compliant UI cards to slide in exactly when the speaker makes the claim.

---

💻 Run it Locally (Docker)

The absolute best way to run VidScrutiny is via Docker, as it automatically handles complex ffmpeg dependencies required for video manipulation.

Prerequisites

• Docker Desktop
• A free Groq API Key

1-Minute Setup

1. Clone the repository to your machine.
2. Create a .env file inside the backend/ directory:

   GROQ_API_KEY=gsk_your_groq_api_key_here

3. Boot the pipeline:

   docker-compose up --build

4. Open http://localhost:8000 in your browser.

---

🌍 Production Deployment Architecture

Because this platform relies on WebSockets and FFmpeg, it uses an industry-standard split deployment:

• The Frontend (Vercel): The frontend/ directory is deployed as a lightning-fast static site on Vercel.
• The Backend (Render): The backend/ directory and Dockerfile are deployed as a persistent Web Service on Render to keep the WebSocket connections alive and execute the heavy video processing.