Translating PDF documents while preserving their complex structure (columns, tables, embedded images) is a notorious engineering bottleneck. Standard tools often destroy the layout or fail to capture text inside images.

I engineered a robust Python automation pipeline that fuses Optical Character Recognition (OCR) with Neural Machine Translation. The system automatically deconstructs PDFs, translates content, and rebuilds the document with the original layout preserved.

Extracts text from PDFs, including images with embedded text (OCR fallback).

Translates text into a target language using a pre-trained MarianMT model.

Recreates the PDF with both the original and translated text for side-by-side comparison.

Skips files that are already translated to avoid redundant processing.

Supports custom fonts for translated text in the recreated PDF.

Logs detailed progress and errors for debugging.

Added a script to generate random sample PDFs and included 5 sample PDFs for demonstration.

Includes logic to skip already-translated files, enhancing efficiency.