Architecture decisions from the systems I build and operate. Real trade-offs, the options I rejected, and what runs in production today.
Oct 2026
I set out to build an AI-native product and had to build Lumo first: a real problem, an MVP that broke on money, a rebuild around a ledger, and three studios of data before AI could mean anything.
Sept 2026Case study
The accounting core of Lumo, taken out as tenant-ledger, a package on npm, to test whether its boundary was real. It was. Reviewing it on its own found an allocation race production had never shown, proved it on two connections, and sent the fix back.
Sept 2026Case study
Multi-tenant scheduling, enrolment, attendance and billing for dance studios, built and operated solo. The architecture, the decisions I had to argue myself out of, and what the first months in production changed, with the numbers.