Architecture Notes
Design decisions, verification results, and engineering philosophy — concrete problems that come up in day-to-day implementation, and the abstract discussions behind them.
Selected articles from the Japanese notes, translated into English.
Structurally eliminating the drift of maintaining the same model across Data Connect, Firestore, Zod, and API DTOs, with code generation from a YAML contract as the single source of truth.
Enforcing multi-repository AI agent work through declared scopes and permissions: per-session guardrail design, a self-restraint that keeps the agent away from its own permission definitions, and a double lock on push.
The design of a CI pipeline that keeps DNS, auth, delivery certificates, and env values — everything left over after config-driven GCP / Firebase provisioning — following the same source of truth.
The design of a platform that chains GCP project creation through Firebase setup, keyless, with a per-service settings file as the source of truth.
A development environment with multiple services sharing one GCP project, presented through two cross-sections: how it runs (runtime) and how it ships (delivery).
The design benefits — and the costs — that come into view when Cloud SQL is the source of truth and Firestore is a read-optimized projection.