Mütercim

Feed a PDF in one language, get the same document in another with the same layout. Mütercim is a command-line pipeline in seven stages, each a folder of JSON you can inspect: cut turns the PDF into page images, layout finds the regions on each page with their bounding boxes, ocr and read extract structured text with a vision model, solve resolves abbreviations and injects a glossary without any API call, translate produces the target languages, and write renders Markdown, LaTeX, PDF or DOCX in the original layout.

The example above is a real page: the scan, then the 43 regions the layout stage found on it — the entries in blue, the footnotes in orange, the column separator dashed — drawn from the tool's own output. It is written in Go and works with any language pair the configured models support; the glossary system is general-purpose, so a domain brings its own terminology files.