Skip to content

PDF Invoice Processing with Google Gemini

In this example, we'll extract structured invoice data (vendor, amount, due date, line items) from PDF documents using Pydantic models. Google Gemini's large 1M+ token context window makes it ideal for processing complex documents with multiple pages and detailed line items.

1. Export your Google API key.

export GOOGLE_API_KEY=<YOUR_KEY>

2. Download a sample invoice to test with.

curl -o sample-invoice.pdf https://github.com/excid3/receipts/raw/main/examples/invoice.pdf

3. Extract a structured Pydantic object from the invoice.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
from toolfront import Document
from pydantic import BaseModel, Field
from datetime import datetime

class Invoice(BaseModel):
    invoice_number: str = Field(..., description="Invoice number or ID")
    vendor_name: str = Field(..., description="Name of the vendor/supplier company")
    client_name: str = Field(..., description="Name of the client/customer")
    total_amount: float = Field(..., gt=0, lt=100000, description="Total amount due in USD")
    due_date: datetime = Field(..., description="Payment due date")
    line_items: list[str] = Field(..., min_items=1, description="List of invoice line items")

# Extract structured data from PDF
doc = Document("sample-invoice.pdf")
invoice: Invoice = doc.ask(
    "Extract all invoice information from this PDF",
    model="anthropic:claude-3-5-sonnet-latest"
)

print(f"Invoice Number: {invoice.invoice_number}")
print(f"Vendor: {invoice.vendor_name}")
print(f"Client: {invoice.client_name}")
print(f"Amount Due: ${invoice.total_amount}")
print(f"Due Date: {invoice.due_date.strftime('%B %d, %Y')}")
print(f"Line Items: {len(invoice.line_items)} items")

>>> Invoice Number: 123
>>> Vendor: Example, LLC
>>> Client: Customer
>>> Amount Due: $19.0
>>> Due Date: March 25, 2024
>>> Line Items: 1 items

Pydantic Fields

Use Field() with descriptions to improve AI extraction accuracy and validators (gt=0, min_items=1) to enforce data quality and business rules.