AI 数据提取——从 PDF、发票和非结构化文本中提取结构化输出

import json
from openai import OpenAI
client = OpenAI(api_key="sk-kn-...", base_url="https://api.kunavo.com/v1")

# The schema is the output contract. Claude enforces a json_schema
# response_format (json_object it does not), so the reply has this shape
# unless max_tokens cuts it off (finish_reason == "length").
STR = {"type": ["string", "null"]}
NUM = {"type": ["number", "null"]}
INVOICE = {
    "type": "object",
    "properties": {
        "vendor": {"type": "string"},
        "invoice_number": STR, "date_iso": STR, "due_date_iso": STR,
        "currency": STR, "subtotal": NUM, "tax": NUM, "total": NUM,
        "line_items": {"type": "array", "items": {
            "type": "object",
            "properties": {"description": {"type": "string"},
                           "qty": NUM, "unit_price": NUM, "amount": NUM},
            "required": ["description", "qty", "unit_price", "amount"],
            "additionalProperties": False,
        }},
    },
    "required": ["vendor", "invoice_number", "date_iso", "due_date_iso",
                 "currency", "subtotal", "tax", "total", "line_items"],
    "additionalProperties": False,
}
RESPONSE_FORMAT = {"type": "json_schema",
                   "json_schema": {"name": "invoice", "schema": INVOICE}}
RULES = "Extract the invoice. Use null for missing fields. Dates in ISO 8601."

# Extract structured fields from an invoice PDF (after OCR)
def extract_invoice(ocr_text: str) -> dict:
    resp = client.chat.completions.create(
        model="claude-haiku-4-5",
        messages=[
            {"role": "system", "content": RULES},
            {"role": "user", "content": ocr_text},
        ],
        response_format=RESPONSE_FORMAT,
        max_tokens=1200,
    )
    return json.loads(resp.choices[0].message.content)

# Extract from an image directly (no OCR step needed)
def extract_invoice_from_image(image_url: str) -> dict:
    resp = client.chat.completions.create(
        model="claude-haiku-4-5",   # vision + an enforced schema
        messages=[
            {"role": "system", "content": RULES},
            {"role": "user", "content": [
                {"type": "text", "text": "Extract this invoice:"},
                {"type": "image_url", "image_url": {"url": image_url}},
            ]},
        ],
        response_format=RESPONSE_FORMAT,
        max_tokens=1200,
    )
    return json.loads(resp.choices[0].message.content)