Computational designer in New York, working in Rhino/Grasshopper, C#, Python and graph neural networks. I build tools that test design claims, and every number I publish traces to the file it was measured from.
sriyathotakura.com · LinkedIn · st3827@columbia.edu
Selected work
- GraphNavigator: a C#/.NET static analyser for Grasshopper definitions that run without errors but give wrong results. Four deterministic checks; 46/46 tests pass. It finds 4 of 4 planted faults on synthetic graphs and 3 of 4 on live Rhino documents, with no false positives.
- LATTICE: a graph neural network (PyTorch Geometric) that stands in for a structural solver on one gridshell family. Axial force R² 0.973 against the direct-stiffness solver on 500 held-out variants. It is a proof of method, not a general solver.
- TRENCH: Cross Bronx Expressway research. Trench geometry from LiDAR, plus 139 community documents indexed for fully local retrieval. The repo marks which prototypes run on synthetic data.
- The Self-Enclosing Line: did the High Line's success build the towers that now shadow it? Sky view factor 0.424, against 0.978 on Chicago's 606.
All thirteen projects are listed in portfolio-projects.