ax-java-agent-optimize

Optimize Java Ax agents by configuring evaluator callbacks that score model behavior, applying AxGEPA and AxBootstrapFewShot to drive improvement within bounded experiment workflows. Use the generated dev.axllm:ax package API to execute repeatable optimization runs with explicit budgets and dataset rows.

GEPA and BootstrapFewShot work together by using evaluator callbacks to score agent behavior and drive optimization decisions. BootstrapFewShot initializes the Java agent with few-shot examples, while GEPA iteratively refines program components based on evaluator feedback within bounded experiment budgets.

Yes, you need the generated dev.axllm:ax package along with AxGEPA, AxBootstrapFewShot, OptimizerEngine, and OptimizerEvaluator support to configure and execute Java agent optimization runs with evaluator-driven improvement and bounded datasets.

Yes, you can persist and reuse optimizer artifacts for repeatable experiments and deployment handoff. The Java Ax optimization workflow supports saving optimizer outputs, allowing you to reproduce runs and maintain consistent agent quality across different environments.

Keep Ax agent optimization runs reproducible by defining explicit budgets and bounded dataset rows before execution. Configure evaluator-driven workflows using the Java package API, and persist optimizer artifacts to ensure consistent results across repeated tuning experiments.

Explicit budgets and dataset rows are required to bound experiment workflows and ensure optimization runs remain repeatable. Defining these constraints allows the OptimizerEngine and OptimizerEvaluator to systematically improve Java agent task quality without uncontrolled resource consumption.