Systems optimization should be part of CI/CD

LEVI is a new algorithmic discovery framework that reduces the cost of AI-driven systems optimization by 3–7× compared to existing approaches, making it feasible to continuously optimize algorithms for specific deployments rather than running expensive one-time optimizations. The framework uses smaller, cheaper language models for most tasks while reserving larger models for critical decisions, and maintains diversity in the search process to avoid convergence on suboptimal solutions. The authors argue that affordable algorithmic optimization should become part of standard CI/CD pipelines, allowing systems to automatically adapt their algorithms as hardware, workloads, and performance requirements change.