KatherLab/LLMAnonymizer-Publication — reverse-engineered prompt

Build me a simple research web app that anonymizes medical reports using a local LLM.

I want to upload PDFs, text files, images, and docx files if Word is available. The app should extract the text, run OCR when needed, and create a downloadable zip with the processed text and PDF files with a text layer. Then I should be able to pick a local llama.cpp compatible model, adjust the prompt, grammar, and temperature, and run a job that finds personal information like names, dates, IDs, and addresses. The result should include a CSV with the original report, the masked report, extracted fields, and metadata.

Please also add a redaction area where I can upload annotation files from Inception, compare them against the anonymized output, calculate basic redaction metrics, and view the original and redacted document side by side. Include fuzzy matching settings so slightly different name spellings can still be found. Keep it practical and clear for research use only.