Modelling a house’s thermal properties and interactions is a difficult task, normally involving surveys and long periods of measurements. I propose a system which can learn all of these properties, or some aggregate representation, dynamically with very little to no input from inhabitants or trained personnel. This means the system would have to use what ever basic monitoring data was available to it and from this data create a model for each room and their interactions. This system must infer where a heating source is, and the way each heating source can affect the house individually and in tandem with multiple other sources. This shall allow for rapid deployment into unknown houses which rather than learning the behaviour of its occupants, learns the behaviour of the building itself.