verification-retry-loop

Verify structured data extracted from an LLM by running deterministic post-checks against a Pydantic model. This process validates the extracted output to ensure high data accuracy before final processing.

Retry LLM extraction with automated feedback when data accuracy checks fail. This approach implements a bounded retry loop, automatically re-running extraction once with verification issues provided as feedback to the model.

Yes, you can use Pydantic models for structured output verification. The Skill supports structured Pydantic extraction from LLMs, applying deterministic checks on the output to validate the extracted data structure.

Automate document processing quality checks by implementing the extract_with_verification function. It runs deterministic post-checks on extracted data and automatically retries once with feedback if issues are detected.

The limitation of a bounded retry loop for LLM data extraction is the maximum retry count of one. If the re-extraction attempt with feedback still fails the deterministic post-checks, the loop stops and does not attempt further retries.