pdf-master-translator
Installation
SKILL.md
PDF Master Translator (V10 Architecture)
This skill provides a battle-tested, "bulletproof" pipeline for translating complex PDF documents. It was forged from extensive trial and error on NASA engineering specifications.
Do NOT attempt to use simple OCR or zero-shot LLM translation for complex engineering documents. They will fail. Use the translator_engine_v10.py script provided in this skill.
Core Capabilities & The V10 Pipeline
This skill relies on a Python script (scripts/translator_engine_v10.py) that implements a specific, multi-agent workflow:
- Layout & Physical Isolation (Masking):
- Never ask an LLM to "ignore the picture and translate the text" on a messy scan.
- The pipeline first detects figures and tables.
- It physically whites out (masks) these regions on a temporary image.
- The "clean" image is sent for translation, eliminating visual hallucinations.
- Original figures are extracted, converted to Base64, and safely appended to the final HTML/PDF.