GEPA

GEPA is an optimization framework that improves prompts, code, and agent configurations by having an LLM read full execution traces (errors, logs, profiling data) and propose targeted fixes through Pareto-efficient evolutionary search, rather than relying on a single scalar reward. It's used by teams at organizations including Shopify, Databricks, Dropbox, and OpenAI, and integrates with frameworks such as DSPy, MLflow, and Pydantic AI as an optimization layer in the agent development stack.