The Evaluator Optimizer is a workflow pattern that uses a generate-evaluate-refine cycle to close quality gaps in an agent’s output. It consists of three components:
- Generator: Produces an initial candidate output.
- Evaluator: Scores the output against a rubric or set of constraints, providing structured critique, not just a pass/fail signal.
- Optimizer/Refiner: The generator revises its output based on the evaluator’s feedback.
This loop continues until a convergence criterion is met (e.g., quality score threshold or iteration cap).
Why it matters: This pattern provides a structured way to handle ambiguity and improve the quality of an agent’s output through iterative refinement.
Connections:
Sources: