An interpretable model that makes predictions by routing inputs through a layer of human-understandable concepts rather than opaque features.