Seconding this… Having both the view model for any device that is interactable (switch, vending machine, button, etc.) as well as the interaction component in scene graph.
This would be a good starting place, and can later be expanded to the input trigger device. And the list of interact prompts on the left side of the screen.