instructor

To extract structured data from unstructured LLM responses, you can use Pydantic schemas to enforce type-safety and validate the output. This approach automatically retries validation errors with actionable feedback to ensure accurate data extraction.

Yes, you can stream partial LLM results for real-time UI updates. Streaming partial results allows applications to display structured data progressively as it is generated and validated against your Pydantic models.

Structured LLM output validation works across multiple providers including OpenAI, Anthropic, and local models. This multi-provider support ensures consistent prompts and Pydantic validation regardless of the underlying language model.

The best way to handle LLM validation failures during data extraction is by implementing automatic retries. When Pydantic schema validation fails, the system provides actionable feedback to the LLM and retries the request to correct the structured output.

To extract multiple nested entities from text, define complex Pydantic models with nested structures. The extraction process handles multi-entity records and validates the entire hierarchical data structure against the defined schema.

LLM structured output may not match your Pydantic schema due to unstructured or invalid responses. Using automatic validation with retries feeds validation errors back to the LLM, forcing it to correct the output to match your schema.