darwinian-evolver

You can automate prompt engineering by running an LLM-driven evolutionary search loop that mutates and evaluates artifacts against a custom fitness function. This structured loop uses trainable and holdout data to optimize prompts iteratively.

Yes, you can evolve regex and SQL queries using the evolutionary loop. The loop applies mutation and evaluation to optimize these artifacts against a fitness function, treating them as organisms that improve over generations.

Yes, you need Python version 3.11 or higher, the uv CLI, and access to an LLM provider via an API key such as OPENROUTER_API_KEY to run the evolution experiments and execute the built-in parrot driver.

The best way to optimize small code snippets is defining a custom Problem that guides organism creation, evaluation, and mutation. The EvolveProblemLoop orchestrates the Evaluator and Mutator to refine code through structured evaluation.

Yes, darwinian-evolver supports model-agnostic evaluation loops. It includes a Parrot and OpenRouter driver with templates that allow you to run evolution experiments across different LLM providers seamlessly.

You should avoid LLM-driven evolutionary search when you lack a definable fitness function or sufficient trainable and holdout data. Without clear evaluation metrics, the mutation and selection loop cannot effectively optimize prompts, regex, SQL, or code artifacts.