Lexis Verify

AI-powered fact-checking platform. Extract every verifiable claim from any document, cross-reference against trusted sources, and get a credibility score backed by evidence.

Features

• Claim extraction — automatically identifies all verifiable factual claims from any input
• Source credibility scoring — rates sources from wire services (Reuters 95/100) to social media (Reddit 20/100)
• Multi-format input — PDF, plain text, article URL, or multiple PDFs at once
• Language detection — works in any language, responds in the document's language
• Comparison mode — upload two documents and find contradictions and agreements between their claims
• PDF report export — download a professional report with all claims, verdicts and sources
• History dashboard — track all past analyses with credibility trends
• REST API — integrate fact-checking into any external system

Verdicts

Verdict	Meaning
TRUE	Supported by credible sources
FALSE	Contradicted by credible sources
PARTIALLY TRUE	Some aspects confirmed, others not
UNVERIFIABLE	Insufficient evidence found

Setup

1. Clone the repo

git clone https://github.com/AnnaGals-10/lexis-verify.git
cd lexis-verify

2. Install dependencies

pip install -r requirements.txt

3. Configure API keys

cp .env.example .env

Edit .env and add your keys:

• OPENAI_API_KEY — from platform.openai.com
• TAVILY_API_KEY — from tavily.com (free tier available)

4. Run the app

python -m streamlit run app.py

API

Run the REST API with:

uvicorn api:app --reload

Endpoints

GET  /                    Health check
POST /verify/text         Verify claims from plain text
POST /verify/pdf          Verify claims from a PDF file

Example — verify text:

curl -X POST http://localhost:8000/verify/text \
  -H "Content-Type: application/json" \
  -d '{"text": "The Eiffel Tower was built in 1889 and is located in Berlin."}'

Response:

{
  "language": "English",
  "n_claims": 2,
  "overall_score": 45,
  "results": [
    {
      "claim": "The Eiffel Tower was built in 1889",
      "verdict": "TRUE",
      "confidence": 99,
      "explanation": "..."
    },
    {
      "claim": "The Eiffel Tower is located in Berlin",
      "verdict": "FALSE",
      "confidence": 98,
      "explanation": "..."
    }
  ]
}

Project structure

fact-checker/
├── app.py              — Streamlit UI
├── extractor.py        — Claim extraction from text, PDF and URLs
├── verifier.py         — Claim verification via Tavily + LLM
├── comparator.py       — Document comparison logic
├── scorer.py           — Source credibility scoring
├── report_generator.py — PDF report export
├── api.py              — FastAPI REST endpoints
├── requirements.txt
├── .env.example
└── uploads/            — Temporary upload directory (gitignored)

Source credibility tiers

Tier	Score	Examples
Highly Trusted	90–99	Reuters, AP, Nature, WHO, CDC
Trusted	75–89	BBC, NYT, Guardian, Economist
Moderate	60–74	Wikipedia, regional outlets
Low	40–59	Unknown .org domains
Unreliable	0–39	Reddit, Facebook, anonymous blogs