PaddlePaddle/PaddleOCR ? reverse-engineered prompt

Build me a Python OCR and document parsing toolkit that can take PDFs, scanned images, phone photos, and common office files, then turn them into clean structured output for AI apps. A user should be able to run it from a simple command line or small service, point it at a file or folder, and get Markdown or JSON with recognized text, tables, document layout, and useful coordinates when available.

It should handle messy real world documents like skewed scans, bad lighting, warped pages, and screenshots. It should work across lots of languages, especially mixed Chinese and English, but not only those.

Please make it lightweight enough to run on CPU, but able to use GPU or other accelerators when present. Include sensible defaults, model download or setup steps, clear examples, and a quick demo that shows extracting a PDF into Markdown for a RAG or chatbot workflow. Look up current PaddleOCR docs online if you need to.