sarvanithin/Buildify

Buildify combines HouseGAN++ (a known generative model for floor plans) with Mixture of Experts routing and IRC compliance validation. The novelty lies in the hybrid MOE+HouseGAN++ architecture and compliance layer, but this is an incremental application of existing techniques to architectural generation. Critical weakness: 1 star, 0 forks, 28 days old, zero velocity suggests this is a personal research project or hackathon submission with no adoption, no community engagement, and no evidence of production maturity. The README claims 'true-to-scale' and 'IRC-compliant' outputs but provides no validation, benchmarks, or user testimonials. The 3D visualization is standard (Three.js pattern). Frontier risk is HIGH because: (1) Generative AI for design is a direct frontier-lab interest (OpenAI/Anthropic exploring multimodal design generation, Google exploring architectural AI); (2) Compliance validation and architectural generation are solvable within a larger design platform; (3) The MOE + GAN combination is publishable but not productionized here. A frontier lab could integrate floor-plan generation as a feature in a broader design assistant or acquire the underlying MOE routing technique without needing this specific repo. No defensibility beyond the novelty of the approach—switching costs are zero, reproducibility is high (standard ML stack), and no data gravity or network effects exist. This is a research prototype, not a defensible product.