Unstructured Output

[collindutter]

• I have read and agree to the contributing guidelines.

Is your feature request related to a problem? Please describe.
Users want some way to get structured output like behavior for non-deterministic output. For instance, if I've got a Rule that says "Talk like a Pirate", I want something that automatically enforces (and automatically corrects) an LLM to output that way.

Describe the solution you'd like

• Use EvalEngine to automatically evaluate output on whether it follows the provided Rulesets.
• Provide a custom function that can do whatever kind of check they'd like.
  Describe alternatives you've considered
  Build the loop myself:

from griptape.engines import EvalEngine
from griptape.structures import Pipeline
from griptape.tasks import PromptTask

engine = EvalEngine(
    criteria="Determine whether the answer is spoken like a pirate.",
    evaluation_params=[EvalEngine.Param.INPUT, EvalEngine.Param.ACTUAL_OUTPUT],
)


pipeline = Pipeline(
    tasks=[
        PromptTask("What is 2 + 2? {% if args and args[0] %}{{ args[0] }}{% endif %}"),
    ]
)


score = -1
reason = None
while score < 0.5:
    pipeline.run(reason)
    score, reason = engine.evaluate(input=pipeline.input, actual_output=pipeline.output)