Homegrown
Machine learning apartment recommendation
Automated space planning has been discussed for decades — physics-based systems, Bayesian analysis, evolutionary algorithms, deep learning — yet architects were still designing apartments manually. Homegrown took a novel approach: instead of generic spatial intelligence, work backwards from the firm's own library of built apartment designs.
The result is a recommendation engine that recognises similarity between spaces regardless of size, shape, mirroring or rotation, delivered as a Revit plugin with zero setup.
Click inside any space and it presents ranked, previously-built layouts, reconstructing the chosen one in seconds with approved 3D content and best-practice metadata. It went viral in the construction-tech space and became a research paper at UCL's first DC I/O conference.
