Lilly Tech Systems
Structured Output
Get reliable, machine-readable output from AI models every time. Learn JSON mode, Pydantic parsing, XML output, and validation strategies that turn unpredictable text into structured data you can trust in production.
What You'll Learn
By the end of this course, you'll know how to extract structured data from any AI model reliably, validate it, and handle edge cases in production.
JSON Mode
Use provider-native JSON modes to guarantee valid JSON output from OpenAI, Anthropic, and Google models.
Pydantic Integration
Define output schemas with Pydantic models and get type-safe, validated structured data from AI responses.
XML Output
When to use XML instead of JSON, and how to extract structured data from XML-formatted model responses.
Validation Strategies
Build robust validation pipelines with retries, fallbacks, and graceful degradation for production reliability.
Course Lessons
Follow the lessons in order for a complete understanding, or jump to any topic.
1. Introduction
Why structured output matters, the challenges of parsing free-text AI responses, and an overview of approaches across providers.
2. JSON Mode
Provider-native JSON modes: OpenAI's response_format, Anthropic's tool-use trick, and Google's response schema. Guaranteed valid JSON.
3. Pydantic Output
Use Pydantic models as output schemas with OpenAI's structured outputs, Instructor library, and custom parsing pipelines.
4. XML Output
When XML beats JSON: multi-part responses, mixed content, and streaming. Parsing XML output from Claude and other models.
5. Validation
Build validation pipelines: schema validation, semantic checks, retry strategies, fallback parsing, and handling partial output.
6. Best Practices
Production patterns: choosing the right format, schema evolution, versioning, monitoring parse failures, and common mistakes.
Prerequisites
What you need before starting this course.
- Basic experience with AI APIs (sending prompts, receiving responses)
- Familiarity with JSON and Python
- Understanding of Pydantic is helpful but not required (we cover it in-depth)