Build a PDF Extraction Pipeline in Python

1. 1
   Set Up Your Environment

   The PDF Extraction Pipeline relies on the `unstructured` Python library and Tesseract for OCR. Install `unstructured` with PDF and Tesseract extras. You must also install the Tesseract engine on your system (e.g., `brew install tesseract` on macOS or via the official installer on Windows).
2. 2
   Extract All Elements from a PDF

   Use the `partition_pdf` function to process a document. It intelligently chooses between text extraction and OCR. The function returns a list of 'Element' objects, such as `Title`, `NarrativeText`, and `Table`.
3. 3
   Force OCR on Scanned Documents

   For image-only PDFs or to ensure OCR is used, set the `strategy` parameter to `"ocr_only"`. The PDF Extraction Pipeline will use Tesseract to read the text from the document images.
4. 4
   Isolate and Extract Tables

   The pipeline automatically detects tables and extracts them as `Table` elements. You can filter the results to process only this structured data. The `.text` attribute contains the content, while `.metadata.text_as_html` provides an HTML representation.
5. 5
   Output as Structured JSON

   For easy integration with other systems, convert the extracted elements into a structured JSON format. This is ideal for feeding data into databases, APIs, or other scripts.

import os
from unstructured.partition.pdf import partition_pdf

# --- Setup ---
# 1. Create a file named 'sample.pdf' in the same directory.
# 2. Make sure you have installed Tesseract: https://tesseract-ocr.github.io/tessdoc/Installation.html
# 3. Run this script: python run_pipeline.py

FILENAME = "sample.pdf"

if not os.path.exists(FILENAME):
    print(f"Error: Please create a PDF file named '{FILENAME}' to run this script.")
else:
    print(f"Processing {FILENAME} with the PDF Extraction Pipeline...")
    # Use the hi_res strategy for better layout analysis and table detection
    elements = partition_pdf(
        filename=FILENAME, 
        strategy="hi_res",
        infer_table_structure=True
    )

    print("\n--- Extracted Elements ---\n")
    # Print the type and first 100 chars of each element found
    for i, element in enumerate(elements):
        element_type = type(element).__name__
        element_text_preview = element.text[:100].replace('\n', ' ')
        print(f"{i+1}. [{element_type}]: {element_text_preview}...")

    print(f"\nTotal elements extracted: {len(elements)}")