Nirnay - Fake news detection in live broadcasts

Overview

Nirnay is an AI-powered pipeline for detecting misinformation in live broadcasts. It extracts text from live video feeds, analyzes claims using NLP, and fact-checks them in real time.

Features

• 🎥 Live Video Processing - Converts YouTube/live video to text using FFMPEG & Whisper.
• 🔎 Misinformation Classification - Uses a DistilBERT-based NLP model.
• ✅ Fact-Checking - Integrates Google Fact Check API.
• 📊 Real-Time Dashboard - Built with Streamlit & Plotly.
• 📝 Summary Generation - Uses GPT/T5 to summarize findings.

Workflow

Below is the workflow for Nirnay, illustrating the process of extracting, analyzing, and fact-checking claims from live video streams.

Installation

1️⃣ Setup Virtual Environment

python3 -m venv ~/env
source ~/env/bin/activate

2️⃣ Install Dependencies

pip install -r requirements.txt

3. cd client and run - npm run dev

4. for the backend, python -m uvicorn run:app --reload

Usage

1. Provide a YouTube Live link or select a sample video.
2. Monitor real-time claims detected in the broadcast.
3. View fact-checked results with verified sources.
4. Summarized insights displayed for quick analysis.

Technologies Used

• Python 3.11
• FFMPEG + Whisper AI (Speech-to-Text)
• Hugging Face Transformers (DistilBERT for claim detection)
• Google Fact Check API (Fact verification)
• Streamlit & Plotly (Dashboard visualization)

Future Enhancements

• 🛠️ Multi-language support for misinformation detection.
• 🚀 Enhanced claim verification using multiple sources.
• 📡 Social media integration for wider tracking.

License

This project is licensed under the MIT License.