	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Anthropic Claude API > Claude PDF and Document Analysis

Claude PDF and Document Analysis

Author: Venkata Sudhakar

Claude can read and analyze PDF documents natively, making it powerful for document-heavy workflows. For ShopMax India, this enables automated processing of supplier invoices, warranty certificates, product specification sheets, and customer complaint letters - all common PDF formats in electronics retail. Claude reads the PDF content directly without requiring separate OCR tools, understanding text, tables, and document structure in a single API call.

PDFs are sent to Claude using the document content block type with source type base64. The media_type must be application/pdf. Claude processes up to 100 pages per document and handles multi-column layouts, tables, and mixed text-image content. For large PDFs, send specific page ranges using the start_page and end_page parameters to reduce token usage. The document content block can be combined with text blocks in the same message to ask specific questions about the document.

The following example shows ShopMax India using Claude to extract structured data from supplier invoices and product specification PDFs:

import anthropic
import base64
import json
import os

client = anthropic.Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY"))

def analyze_pdf(pdf_path: str, question: str) -> str:
    with open(pdf_path, "rb") as f:
        pdf_data = base64.standard_b64encode(f.read()).decode("utf-8")
    response = client.messages.create(
        model="claude-opus-4-5",
        max_tokens=800,
        messages=[
            {
                "role": "user",
                "content": [
                    {
                        "type": "document",
                        "source": {
                            "type": "base64",
                            "media_type": "application/pdf",
                            "data": pdf_data
                        }
                    },
                    {
                        "type": "text",
                        "text": question
                    }
                ]
            }
        ]
    )
    return response.content[0].text

def extract_invoice_data(pdf_path: str) -> dict:
    question = (
        "Extract the following fields from this supplier invoice as JSON: "
        "invoice_number, supplier_name, invoice_date, total_amount, currency, "
        "line_items (list of product, quantity, unit_price). "
        "Return only valid JSON, no explanation."
    )
    result = analyze_pdf(pdf_path, question)
    try:
        start = result.find("{")
        end = result.rfind("}") + 1
        return json.loads(result[start:end])
    except Exception:
        return {"error": "Could not parse invoice", "raw": result[:200]}

def summarize_spec_sheet(pdf_path: str, product_name: str) -> str:
    question = (
        "Summarize the key technical specifications for " + product_name + " "
        "from this spec sheet. Format as: Model, Key specs (3-4 bullet points), "
        "Target use case. Keep it under 80 words for ShopMax India catalog use."
    )
    return analyze_pdf(pdf_path, question)

invoice_data = extract_invoice_data("samsung_invoice_may2026.pdf")
print("Extracted Invoice Data:")
print(json.dumps(invoice_data, indent=2))
print()
spec_summary = summarize_spec_sheet("lg_ac_spec_sheet.pdf", "LG 1.5T Dual Inverter AC")
print("Spec Sheet Summary:")
print(spec_summary)

It gives the following output,

Extracted Invoice Data:
{
  "invoice_number": "SSI-2026-04821",
  "supplier_name": "Samsung India Electronics Pvt Ltd",
  "invoice_date": "2026-05-02",
  "total_amount": 549900.00,
  "currency": "INR",
  "line_items": [
    {"product": "Samsung 55-inch 4K QLED QA55Q70D", "quantity": 10, "unit_price": 54990}
  ]
}

Spec Sheet Summary:
Model: LG 1.5T Dual Inverter AC (S-Plus 5-Star)
Key specs:
- Dual inverter compressor with 70% energy savings
- Auto cleaning and HD filter for air quality
- 4-way swing and monsoon comfort mode
Target use case: Energy-conscious Indian homes needing powerful cooling with
low electricity bills in humid cities like Mumbai and Chennai.

For ShopMax India production document pipelines, upload frequently used PDFs (standard supplier contracts, warranty templates) to the Files API once and reuse the file_id across multiple requests - this avoids re-uploading the same PDF on every call and enables prompt caching of the document content. When processing large batches of invoices, combine Claude PDF analysis with the Batch API to process hundreds of documents overnight at 50% cost. Always validate extracted JSON against a Pydantic schema before inserting into your database - Claude occasionally misses fields in complex multi-page invoices.

Send your comments, suggestions or queries regarding this site to [email protected].