ce-optimize

Metric-driven optimization loops require an optimization spec YAML file to define measurable goals, run parallel experiments, and evaluate clustering quality using hard gates or LLM-as-judge scores to keep improvements.

LLM-as-judge evaluation scores experiment outputs against defined metrics in your optimization spec, systematically measuring prompt quality or clustering relevance to keep only proven improvements during iterative experimentation.

The best way to automate parallel experiments is using an optimization spec YAML to define parameters and a persistent experiment log schema to track improvements, converging on the best solution through systematic measurement.

Yes, you need an optimization spec YAML file to define the metric, hard gates, judge configuration, and other parameters required to run metric-driven iterative optimization loops and measure experiment success.

Yes, you can optimize search relevance and prompt quality by defining measurable goals and hard gates in the optimization spec YAML, running parallel experiments to systematically evaluate and keep improvements.

Systematic experimentation converges on the best solution by running parallel experiments and measuring each against hard gates or LLM-as-judge scores, avoiding the guesswork and inconsistency of manual parameter tuning.