javvi51/edocr2 ? reverse-engineered prompt

Build me a Python tool that can take engineering drawings, especially mechanical or production drawings, and find the text and drawing regions, then run OCR on the text it finds.

I want it to be something I can run on sample drawings to quickly test the results, and I also want a way to train custom recognizers or detectors with synthetic data. Please include a simple way to benchmark it on a folder of test drawings, and another small example that shows how it can work with an LLM if I want to send OCR results into one later.

Make it easy to install and run from the command line, with clear example scripts I can try right away. If you need current library details or model download steps, look them up online and wire it up in the simplest way that works.