ai-agents-structured-output-re-structured-llm-output-fails-to-parse-into-pydantic-604f2571

THEN Implement a retry mechanism that catches Pydantic validation errors and re-calls the LLM with the same prompt, optionally including the error message as feedback. This ensures the output conforms to the expected schema before proceeding, similar to the retry logic used in the ReflectionWorkflow.

from pydantic import ValidationError

max_retries = 3
for attempt in range(max_retries):
    response = sllm.chat(messages)
    if isinstance(response, str):
        # parse error, retry
        continue
    try:
        output = response.model_dump_json()
        break
    except ValidationError as e:
        # optionally include error feedback in next prompt
        continue