Data Sanitization and Knowledge Distillation

Problem

Business documents often contain names, companies, amounts, and project details. Uploading or reusing them directly can expose identifiable entities, while raw notes often remain case-specific and hard to reuse.

How It Works

• Supports .txt, .md, .csv, .docx, .pdf, .pptx, or pasted text; non-plain files are first extracted to .md.
• Runs scripts/anonymize.py for regex-based sanitization, then uses the LLM to handle English names, contextual references, and combined-identity leaks.
• Produces [filename]_anonymized.md to preserve narrative and [filename]_knowledge.md to distill underlying logic, trigger conditions, SOPs, or decision trees.

Notes

• Sanitization takes priority over readability; ambiguous information is masked by default.
• The knowledge document must be transferable, not just a retelling of one case.
• The original file is not modified; anonymized output is Markdown and does not preserve fonts, colors, or other styling.