ce-optimize

Metric-driven iterative optimization executes sequential experiments to improve a measurable goal by testing variations, evaluating them against hard gates or LLM-as-judge scores, and converging on the best solution. You simply provide an optimization spec or a goal description to start the loop.

Parallel execution accelerates optimization loops by running multiple experiments simultaneously. This allows the system to test more variations concurrently, measure each against defined metrics, and rapidly converge on the optimal solution without waiting for sequential trials to finish.

Systematic iterative experimentation applies to any measurable outcome, such as optimizing clustering quality, search relevance, build performance, or prompt quality. If your workflow benefits from metric-driven improvement and convergence, it can be optimized using this approach.

To start an optimization loop, you need to provide either a valid optimization spec YAML file or a text description of your optimization goal. This input defines the measurable target and parameters for the parallel experiments to run.

Experiment variations are evaluated by measuring each against hard gates or LLM-as-judge scores. The system keeps the improvements that pass these metrics and discards the rest, ensuring the loop converges systematically toward the best solution.

Optimization experiment results are saved for durability and analysis through built-in results persistence. This ensures that all data from your parallel experiments and convergence metrics are securely stored for future review.