instructor

You extract structured data from LLMs by using Pydantic schemas to define expected outputs and applying automatic validation loops to parse and enforce type safety on the generated text.

The best way to get consistent JSON output from OpenAI and Anthropic models is to use structured data extraction with Pydantic validation, which automatically feeds parsing errors back to the LLM for correction via retries.

Handle Pydantic validation errors by enabling automatic retries, which feed the specific validation failures back to the LLM so the model can correct its output and satisfy the schema requirements.

Yes, you can stream structured data extraction results in real-time to process partial outputs incrementally, improving user experience while maintaining Pydantic schema validation for type safety.

Yes, structured data parsing works with both OpenAI and Anthropic, allowing you to extract complex multi-entity data and user profiles across diverse LLM providers using Pydantic schemas.