Verifact

Smart AI-powered system designed to analyze news headlines, paragraphs, image snapshots, and URLs to mitigate misinformation. It thoroughly evaluates claims against verified fact-checks and live web search data to deliver a concrete authenticity verdict.

---

Key Features

• Multi-modal Input: Accepts Live News URLs, Image Uploads (news screenshots/posters) and direct Text/Headline inputs.
• Intelligent Claim Extraction: Identifies the core, verifiable claims within the provided information.
• Hybrid Verification: Checks claims against pre-ingested fact-checks and performs live web searches for the latest context.
• Transparent Reasoning: Outputs a definitive verdict (True, False, Misleading, Unverified) with a confidence score, detailed reasoning, claim-by-claim breakdowns and the exact source URLs consulted.

---

Tech Stack

• Backend: FastAPI (Python)
• Frontend: HTML, CSS3, and JavaScript.
• Agent Framework: LangChain and LangGraph.
• LLMs: Open-weight and proprietary models via multiple inference providers.
• Information Retrieval:
  • Pinecone
  • HuggingFace sentence-transformers
  • Tavily AI for Web Search
• Data Scraping & OCR:
  • EasyOCR & pytesseract
  • Newspaper3k & BeautifulSoup

---

Architecture Workflow

       Input (Image / URL / Text)
                   │
                   ▼
             Input Router
             ┌─────┴─────┐
             │           │
          OCR Tool   URL Scraper
             └─────┬─────┘
                   │  extracted text
                   ▼
         Claim Extractor (LLM)
                   │
                   ▼
              Agent Graph
             ┌─────┴─────┐
             │           │
         RAG Search  Web Search
             └─────┬─────┘
                   │  aggregated evidence
                   ▼
           Verdict Synthesizer
                   │
                   ▼
       Structured Output (JSON)
 { verdict, confidence_score, claims_analyzed, sources }

---

Project Structure

├── app/
│   ├── agent/       
│   ├── api/           
│   ├── models/       
│   ├── multimodal/   
│   ├── rag/           
│   ├── config.py      
│   └── main.py        
├── frontend/
│   └── static/        
├── data/
│   ├── raw/                    
├── Dockerfile         
└── README.md

---

How to run locally

1. Clone the Repository:

   git clone https://github.com/your-username/verifact.git
   cd verifact

2. Set up Environment and Google CLI for Vertex AI access: Create a .env file from the example:

   cp .env.example .env
   

3. Install Dependencies:

   uv pip install -r requirements.txt

4. Start the Server:

   uvicorn app.main:app --reload

   Access the UI at: http://127.0.0.1:8000

---

License

This codebase is released under the Apache 2.0 License.