the project’s base reference towards developing machine learning tools for the generation of environmentally reactive floorplans was a catalogue of the distribution and shape of heat islands within the existing urban context. the plan_gan was trained as a pix2pix network with increasingly complex architectural floorplans and the respective environmental simulations of their surrounding. in the final workflow, the plan_gan is evaluated at various sectional heights on a site, according to which a continuous 3D structural and spatial configuration is generated. this architectural output aims to actively counteract a site's urban heat islands through geometric and functional considerations.

the algorithm aims to provide production viability to the generated complex geometries through an incorporated tiling system. the larger structure automatically gets broken into smaller building blocks, which are optimized structurally for off-site precast production. necessary substructure and connection details are also generated within this computational step.

the final output becomes a proof on concept for the revitalization of neglected city streets and the development of new typologies of public space that manage to unite different areas of private and public in close proximity. this proposal, parasitically inhabiting the facades of existing buildings, aims to blur the line between human intuition and machine intelligence, ultimately creating spaces that merge natural and artificial urban landscapes into a vibrant collection of thresholds.

the spatial success of the workflow was measured in a final step through a series of large-scale resin prints. the physical models showcased a plurality of unexpected sectional relationships between different spaces of the intervention.