on a technical level, the system utilizes a bottom-up software interface where a reinforcement learning agent provides live feedback during the architectural formfinding process. the methodology begins by cataloguing 3d-scanned, deconstructed timber members into a structural database, sorting them by physical dimensions. a custom aggregation algorithm then recursively subdivides geometry to assemble these non-standardized pieces into viable laminated beams, utilizing machine intelligence to manage material scarcity and structural complexity.

at the urban scale, the methodology drives a data-driven masterplan for the relocation of kiruna, sweden, a city undermined by intensive mining activities. instead of relying on static zoning, the model cross-references dynamic geological stability metrics and civic requirements to suggest adaptive growth patterns. image-based pix2pix models are trained to translate these large-scale zonings into volumetric space divisions and floorplan suggestions, allowing planners to dynamically adjust the city's layout to changing physical and industrial realities.

the architectural resolution translates these computational protocols into spaces that address sustainability, destruction, and cultural preservation under violent environmental changes. by 3d scanning physical spaces of meaning in the old city, familiar architectural artifacts are captured, distorted, and mapped onto the facades of the new structures. the resulting architecture acts as a physical and cultural archive, superimposing past historic heritage onto a flexible, resource-scarce production model.