How to Extract Structured Data from PDFs with AI

For invoices, receipts, and unstructured documents, AI is the most accurate approach:

from openai import OpenAI
import fitz
import json

client = OpenAI()

def ai_extract(pdf_path, extraction_type="invoice"):
    # Extract text from PDF
    doc = fitz.open(pdf_path)
    text = "\n".join(page.get_text() for page in doc)
    doc.close()
    
    prompts = {
        "invoice": "Extract: invoice number, date, vendor, line items (description, qty, price, total), subtotal, tax, grand total. Return JSON.",
        "receipt": "Extract: store name, date, items (name, price), subtotal, tax, total, payment method. Return JSON.",
        "resume": "Extract: name, email, phone, skills, work experience, education. Return JSON."
    }
    
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {"role": "system", "content": "Extract structured data. Return valid JSON only."},
            {"role": "user", "content": f"{prompts[extraction_type]}\n\nDocument:\n{text[:30000]}"}
        ],
        response_format={"type": "json_object"},
        temperature=0.1
    )
    
    return json.loads(response.choices[0].message.content)

# Extract invoice data
data = ai_extract("invoice.pdf", "invoice")
print(json.dumps(data, indent=2))

The response_format={"type": "json_object"} ensures the output is always valid JSON, avoiding parsing errors.

Once you have structured data, convert it to a spreadsheet:

import pandas as pd

def to_excel(extracted_data, output_path):
    # For invoice line items
    if "line_items" in extracted_data:
        df = pd.DataFrame(extracted_data["line_items"])
        
        # Add summary row
        summary = pd.DataFrame([{
            "description": "TOTAL",
            "total": extracted_data.get("grand_total", "")
        }])
        df = pd.concat([df, summary], ignore_index=True)
        
        df.to_excel(output_path, index=False)
        print(f"Saved to {output_path}")

to_excel(data, "invoice_data.xlsx")

Or skip all the code and use Reformat's PDF to Excel or AI Data Extractor tool — upload your PDF and download a clean spreadsheet in seconds.

For processing many documents at once:

import os
from pathlib import Path

def batch_extract(input_dir, output_dir, extraction_type="invoice"):
    Path(output_dir).mkdir(exist_ok=True)
    
    pdfs = [f for f in os.listdir(input_dir) if f.endswith(".pdf")]
    results = []
    
    for pdf_file in pdfs:
        pdf_path = os.path.join(input_dir, pdf_file)
        try:
            data = ai_extract(pdf_path, extraction_type)
            data["source_file"] = pdf_file
            results.append(data)
            print(f"Extracted: {pdf_file}")
        except Exception as e:
            print(f"Failed: {pdf_file} — {e}")
    
    # Save all results as one JSON file
    with open(os.path.join(output_dir, "all_extracted.json"), "w") as f:
        json.dump(results, f, indent=2)
    
    print(f"Processed {len(results)}/{len(pdfs)} files")

batch_extract("./invoices/", "./extracted/", "invoice")

This processes all PDFs in a directory and outputs a single JSON file with all extracted data. At gpt-4o-mini prices, processing 100 invoices costs about $0.50.