wxyhgk/retain-pdf — reverse-engineered prompt

Build me a usable PDF translation tool that keeps the original page layout as much as possible. I want it to work well for research papers, technical documents, books, and especially scanned or image based PDFs, not just clean copyable ones. It should do OCR when needed, preserve formulas and document structure, avoid mistranslating code snippets, and let me control how tables are handled. The translated result should still look close to the source, with sensible font sizing and a reasonably small output PDF.

Please make it feel like a complete product, not just a script. I want a simple browser interface as the main entry point, a desktop app build if possible, and a Docker deployment option for team or local network use. Include upload, task progress, file library, output artifacts, and basic resume or recovery for interrupted jobs. It would also be great to support configurable translation rules or glossary behavior and an API for automation. Look up current docs in the repo if you need to.