Phase 5: Modern LLM Engineering · 70 min · OpenAI Structured Outputs · Instructor · Pydantic

Structured Outputs & Reliable Extraction

If your LLM output cannot be parsed by code, it is not production-ready.

Hiring signal: Reliable data extraction from unstructured text at scale

The Problem

You build an invoice processor that extracts vendor name, amount, date, and line items. It works 90% of the time. The other 10%, the model returns slightly different JSON keys, adds commentary outside the JSON, or hallucinates a field. Your downstream code crashes. Users lose trust. The product cannot scale.

Structured outputs solve this: they guarantee the model's response conforms to a predefined schema, every time.

The Concept

There are three levels of structured output reliability:

You need the LLM to return a JSON object with exact field names and types for your API to parse. "Please respond in JSON" gives ~80% reliability. What should you use instead?

Schema-enforced outputs (Level 3) constrain the model's decoding to only produce tokens that match the JSON Schema. The API guarantees valid, parseable JSON with correct field names and types. This eliminates the ~20% failure rate of prompt-only JSON, removing the need for retry logic or regex fallbacks.

How Schema-Enforced Outputs Work

When to Use Each Approach

Approach	Use When	Limitation
Structured Outputs (OpenAI)	You need guaranteed schema compliance	OpenAI-specific
Instructor library	You want Pydantic models with any provider	Adds retry logic
Outlines	You need schema enforcement on local models	Requires model access
JSON mode	Schema is simple and retries are acceptable	No schema guarantee
Prompt-only JSON	Quick prototypes only	Frequently breaks

Unlock the full lesson

You've read the first 2 sections. The rest of this lesson covers Build It, Use It, Ship It, Evaluation, Common Pitfalls, Exercises, Key Terms, Interview Framing — plus a hands-on lab, quiz, and project artifact.

Create a free account to unlock Phase 0 and Phase 1 of every course — no credit card.

Key terms in this lesson

Function Calling