AI Citation Auditor 📚🔍

An automated tool that uses Google Gemini (Flash & Pro) to verify academic citations in your manuscripts. It checks if citations exist, if the metadata is correct, and if the cited paper actually supports the claims made in your text.

Features

• Dual Mode Support:
  • docx_citation_checker.py: Scans Microsoft Word (.docx) files directly.
  • tex_citation_checker.py: Scans LaTeX (.tex) files and parses standard .bib files.
• AI-Powered Verification:
  • Uses Gemini Flash for rapid, low-cost checking.
  • Automatically escalates "Suspicious" or "Partial" matches to Gemini Pro for deep reasoning.
• Smart Context: Understands "split support" (e.g., when multiple citations (1, 2) collectively support a sentence).
• Audit Reports: Generates a detailed CSV audit log and a "Deviant Entries" text report highlighting only the problematic citations.

Setup

1. Clone the repository:

   git clone https://github.com/SahPet/ai-citation-auditor.git
   cd ai-citation-auditor

2. Install requirements:

   pip install -r requirements.txt

3. Get a Gemini API Key:

   • Get a free key from Google AI Studio.
   • Create a file named .env in the project folder:

   GEMINI_API_KEY=your_actual_api_key_here

Usage

For Word Documents (.docx)

1. Run the script:

   python docx_citation_checker.py

2. Select your .docx file in the popup dialog.

3. The script will generate:

   • filename_audit.csv: Full report of all citations.
   • filename_deviant_entries.txt: Detailed reasoning for citations marked "Incorrect" or "No".

For LaTeX Projects (.tex + .bib)

1. Run the script:

   python tex_citation_checker.py

2. Select your main .tex file and your .bib bibliography file when prompted.

3. Review the generated console output and CSV reports.

How It Works

The tool uses a two-stage verification process:

1. Extraction: It parses your document to find every citation and the sentence it belongs to (the "Context").
2. Flash Check: It sends the Citation + Context to Gemini Flash to ask: "Does this paper support this claim?"
3. Pro Audit: If Flash returns "No", "Partially", or "Suspicious Integrity", the tool calls Gemini Pro to double-check, providing a detailed "Why is this better?" reasoning if a replacement is suggested.

License

MIT License