HUD is a platform for building reinforcement learning environments that encode human expertise into evaluations, training tasks, and post-training data. Its system includes an SDK for defining evals, environments, and verifiers, plus a training and evaluation platform for running models on those environments. The company reports 1.3M task runs, 2,500 environments created, and 15M inference calls, with customer stories from DoorDash, Sharpe, and UiPath.