Drove evaluation infrastructure and performance and accuracy improvements for AWS DevOps AI agents, establishing robust benchmarks and identifying key areas for model and system-level gains.