Validate LLM Output with Pydantic

Trust, but verify

Even with JSON mode, validate before you use the data. Pydantic turns a schema into a parser that fails loudly on bad output.

from pydantic import BaseModel, ValidationError

class Article(BaseModel):
    title: str
    tags: list[str]
    minutes: int

try:
    article = Article.model_validate_json(llm_response)
except ValidationError as e:
    # retry with the error appended to the prompt
    ...

The pattern

Define the schema, parse the response, and on failure retry once with the validation error fed back to the model. This simple guard makes LLM output safe to build on.