Structured extraction

Turning unstructured text into JSON your code can rely on.

Asking for JSON

response_format is a supported request field. Setting it to {"type": "json_object"} asks the model to emit valid JSON.

Python

import json, os
from openai import OpenAI

client = OpenAI(base_url="https://api.astrodyne.ai/v1",
                api_key=os.environ["ASTRODYNE_API_KEY"])

resp = client.chat.completions.create(
    model="YOUR_MODEL_ID",
    response_format={"type": "json_object"},
    messages=[
        {"role": "system",
         "content": "Extract fields as JSON with keys: name, email, company."},
        {"role": "user", "content": raw_text},
    ],
)

data = json.loads(resp.choices[0].message.content)

Validate what comes back

Asking for JSON is not a guarantee of your schema. Parse it, validate it against a schema you control, and decide what to do when a field is missing. Treat model output as untrusted input, exactly like a form submission.

Making extraction reliable

Name every field you want, in the system message, with its type.
Say explicitly what to emit when a value is absent — null beats invention.
Set a low temperature. Extraction is not a creative task.
Use seed when you need repeatable output for a test.

Cost shape

Extraction is input-heavy and output-light, so the input price per million tokens dominates. An efficient-tier model is often the right choice; compare in Models.

Chat Completions → Every supported field. Limits → Controlling batch spend.