Structured Outputs

Loose JSON mode can return wrong keys or missing fields. Schema enforcement eliminates an entire class of validation bugs in agent pipelines.

Structured outputs are a customs form with fixed boxes — name here, date there — not a blank page labeled 'JSON please.'

Scenario

A ticket classifier must return {priority: 'high'|'medium'|'low', category: string, summary: string} — strict schema ensures no missing priority field.

Solution

In Tool Calling & Function Calling, apply Structured Outputs to this scenario: A ticket classifier must return {priority: 'high'|'medium'|'low', category: string, summary: string} — strict schema ensures no missing priority field. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Structured Outputs (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Structured Outputs happens in the code.

1from pydantic import BaseModel  # import dependencies2class Ticket(BaseModel):  # define a data structure or component3    priority: str4    category: str5    summary: str6
7response = client.beta.chat.completions.parse(8    model="gpt-4o-mini",9    messages=[{"role": "user", "content": ticket_text}],10    response_format=Ticket,11)12ticket = response.choices[0].message.parsed

• Treating Structured Outputs as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for structured outputs

Quick recap — the most important points from this module.