document-ocr

Use pymupdf to open the PDF and call get_text on each page, joining the results into a single string. This works instantly for text-based PDFs but returns empty output for scanned documents, which require OCR instead.

Use liteparse via its lit CLI with the parse command to run OCR on scanned PDFs and images, outputting markdown-formatted text. If liteparse is unavailable, a vision model can serve as a fallback extraction method.

pymupdf is fast and ideal for born-digital PDFs with embedded text layers. liteparse performs OCR, making it necessary for scanned PDFs and image files where no text layer exists.

Empty output usually means the PDF is a scanned image without a text layer, so pymupdf finds nothing to extract. Fall back to an OCR engine like liteparse, and always validate that extracted text is non-empty before returning it.

Yes, but the file must be downloaded locally first using a Google Drive API tool, then the local path is passed to the extraction function. The OCR step itself only accepts local file paths or file bytes.