Platform
Platform work is the part nobody demos: the repo scaffold, the identity boundary, the promotion gate, the registry that makes a model reproducible six months later. I build the factory first so every model after it costs a fraction of the first — infrastructure as code, environments that differ only by variable, and a reference model kept alive end-to-end as proof the path still works.
Tokyo Eye
Platform lens — isolated lab packaging, pinned scientific dependencies, and reproducible runs so hyperbolic and MoE experiments can be re-run, not just reported.
AI/ML Ops Factory
Platform lens — the reference implementation: AIML-SCAFFOLD generates Terraform project repos with OIDC-authenticated CI and promotion gates; a separate live taxi path trained and registered on Azure ML.