script/ocr.py

import os
from pdf2image import convert_from_path 
import pytesseract
from langchain_core.documents import Document
from script.errors import IngestionError


def extract_text_via_ocr(pdf_path: str) -> list[Document]:
  POPPLER_PATH = os.getenv("POPPLER_PATH", "/usr/bin")

  images = convert_from_path(
     pdf_path,
     poppler_path = POPPLER_PATH

  )
    
  documents = []

  for i, img in enumerate(images):
      text = pytesseract.image_to_string(img)
      if text.strip():
          documents.append(
              Document(
                  page_content=text,
                  metadata={
                      "page": i,
                      "source": "ocr"
                  }
              )
          )

  if not documents:
        raise IngestionError(
            code="OCR_FAILED",
            message="OCR produced no text"
        )

  return documents