Introduction
The building and urban sectors account for a major share of global energy use, yet current energy-efficient design remains constrained by linear workflows and limited design-space exploration. This paper proposes an AI-driven parametric design framework that couples generative form-finding with performance-based evaluation to support the creation of ultra-low-energy buildings and districts. The focus is on retaining architectural authorship while leveraging artificial intelligence to navigate complex trade-offs between form, comfort, and energy demand.

Methods
A multi-stage workflow is developed using a parametric modelling environment linked to energy simulation and AI-based optimisation. Parametric building and district models encode key geometric and material variables (massing, orientation, envelope articulation, glazing ratios, and passive systems). Surrogate models trained on simulation datasets (e.g., regression and neural networks) predict energy and comfort metrics in real time. Multi-objective genetic algorithms and reinforcement learning agents generate and refine candidate configurations, while an architect-in-the-loop interface allows designers to steer and constrain the search.

Results
Applied to prototypical mid-rise building typologies and mixed-use district scenarios, the workflow rapidly produces large families of design variants and visualises Pareto fronts between energy use intensity, daylight availability, and floor area efficiency. Compared to conventional iterative design, the AI-enhanced approach uncovers solutions with substantially lower predicted annual energy demand while preserving spatial quality and façade diversity. At the district scale, coordinated optimisation of building forms and orientations demonstrates improved load smoothing and reduced peak demands.

Conclusions
The study shows that AI-driven parametric architectures can transform energy-oriented design from a late-stage validation exercise into an integrated generative driver of form at building and urban scales. The proposed workflow reinforces, rather than replaces, architectural judgement, and suggests a path towards explainable, performance-informed design cultures for ultra-low-energy built environments.