Gianluca Rizzo
First Name: Gianluca
Last Name: Rizzo
Bio
Accepted Talks:
LLMs as Mutation Operators: Building an Evolutionary Search Harness in Python
Evolutionary optimisation offers an alternative method to discover high-performing algorithms by making use of LLMs. It does this by mimicking natural evolution: maintaining a population of plausible solutions, and continuously combining them in new ways to produce better ones. The LLM acts as an intelligent mutation and crossover operator inside the automated search loop, systematically exploring the problem space and driving discovery towards novel, high-performing solutions.
In this session, we walk through building a lightweight evolutionary harness from scratch in Python using Pydantic AI. We break down the key technical components, from tournament selection and LLM-driven code mutation, to robust and automated evaluation mechanisms. Beyond setting up the core evolutionary loop, we also explore engineering considerations for such a system to exist in production.
Using the code we build together, you will walk away with a practical framework for using evolutionary LLM search in your own Python projects.
