Build a scalable replayable-environment engine from enterprise historical data to train, evaluate, and improve agents. Own environment factory, reward/grade design, task mining, eval set creation, agent training/evaluation, replay/observability, and production-scale infrastructure.