Skip to main content

The Vector Loss-Free Grounding Principle

[!IMPORTANT] Mandatory PDF Pre-Processing Policy: Word (.docx, .doc), PowerPoint (.pptx, .ppt), and Hangul (.hwp) files must be converted (baked) to PDF (.pdf) prior to uploading to AXON.

Why Are Raw Office Files Restricted?

  1. Parser Layout Skew: Proprietary office XML formats frequently experience dropped tables, misplaced footnotes, and font substitution during server-side text extraction.
  2. Hallucination Vectors: Missing table cells or misplaced footnotes in a financial or legal report cause AI models to hallucinate figures.
  3. Exact Visual OCR Grounding: Exporting to PDF preserves the fixed vector layout, exact coordinates, and typography. AXON’s multimodal vision engine indexes the exact visual page, guaranteeing 100% loss-free grounding and citation accuracy.

How to Prepare Files for Upload

  • Microsoft Word / PowerPoint: Use File > Save As / Export > PDF.
  • Hancom Office (HWP): Use File > Save As PDF.
  • Google Docs / Slides: Use File > Download > PDF Document (.pdf).