This script uses Google's Document AI and the OpenAI API to go beyond simple extraction. It reads an invoice, extracts key data, and then uses GPT-4 to identify any unusual payment terms.

Program begins

# Requires: pip install google-cloud-documentai openai python-dotenvimport osfrom google.cloud import documentaifrom openai import OpenAIfrom dotenv import load_dotenvload_dotenv()  # Load API keys from .env file# Configure clientsdocai_client = documentai.DocumentProcessorServiceClient()openai_client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))# 1. Process Document with DocAIPROJECT_ID = "your-google-cloud-project-id"LOCATION = "us" PROCESSOR_ID = "your-invoice-processor-id" file_path = "invoice.pdf"with open(file_path, "rb") as image:    image_content = image.read()document = {"content": image_content, "mime_type": "application/pdf"}name = f"projects/{PROJECT_ID}/locations/{LOCATION}/processors/{PROCESSOR_ID}"request = {"name": name, "raw_document": document}result = docai_client.process_document(request=request)document_json = result.document.to_json()# 2. Extract structured data from DocAI response (simplified)# In a real scenario, you would parse the full entity list from document_jsoninvoice_data = {    "supplier_name": "ABC Supplies Ltd.",    "invoice_total": "$12,345.67",    "due_date": "2023-12-01",    "payment_terms": "Net 90"  # Let's say this is unusual}# 3. Use Generative AI for Analysis & Anomaly Detectionprompt = f"""Analyze the following invoice data and point out any unusual or non-standard terms for a typical B2B transaction.Focus on payment terms, large amounts, or missing information.Invoice Data:{invoice_data}Provide a concise summary for an accounts payable specialist."""response = openai_client.chat.completions.create(    model="gpt-4",    messages=[        {"role": "system", "content": "You are a helpful financial analyst assistant."},        {"role": "user", "content": prompt}    ],    max_tokens=200)analysis = response.choices[0].message.contentprint("*** Generative AI Analysis ***")print(analysis)# Output might be: "Warning: The payment terms 'Net 90' are unusually long for this industry. Standard terms are typically Net 30. The invoice total of $12,345.67 is high and may require managerial approval according to company policy."