dewdew978/isd-2026-luksuitpiti ? reverse-engineered prompt

Build me a Python OCR tool for Thai and English documents that can read both single images and multi page PDFs. I want to be able to drop files into an input folder, run one command from the terminal, and get the extracted text saved into an outputs folder as both a clean text file and a JSON file with details like page number and confidence.

Please support a few OCR options, including PaddleOCR, Tesseract with Thai and English, TrOCR for short cropped text, and a simple combined mode that can compare results from more than one engine. It should also preprocess the image first, like resizing, denoising, deskewing, and thresholding, so scanned documents work better.

I also need a way to compare OCR results against ground truth and get useful evaluation scores like CER and WER. If possible, extract common fields like dates, email, phone number, and ID numbers from the OCR text. Make it easy to run from a CLI, and if you need current docs for any library, look them up online.