How would you handle Pydantic output_type validation and retries?

[jackien1]

If you want to double check that a property like "pdf_id" that the AI spits out in structured output exists in the database would you just run a while loop checker with max iterations, dynamic model creation of enums or literals, or use AfterValidator? Not sure what is the best practice for this agentic SDK.

[rm-openai]

I think the most straightforward thing would be to take the final output, so the validation, and it the validation failed - re-run the agent with the previous input list plus a new message saying that the output was invalid. Something like "pdf didn't exist, please fix"