The same platform at four maturity stages. Stage 1 is an MVP you ship in a week
(Vercel + Cloud Run
+ Cloud SQL — the smallest proven stack). Advance the stage to
add only what that stage needs; future components stay dimmed so you see the roadmap without the
overwhelm. This is the antidote to over-engineering: nothing is added before its time.
One architecture, revealed stage by stage (click any box for its definition)
active at this stage arrives in a later stage
Stage 1 · MVP (ship it). The smallest proven stack: a Vercel
frontend, one Cloud Run service, one
Cloud SQL database. You can demo quote→bind end-to-end.
Add nothing else yet.
Stage 2 · Real product. Add what makes it a real insurance product: a
product-version store (versioned by date), the
external cat-model for peril scores,
Pub/Sub for the scorecard fan-out, and a
WAF at the edge (quote endpoints get scraped).
Stage 3 · Trust & AI. Now earn trust: make the lifecycle
event-sourced with a hash-chained
audit log + WORM
anchor; add the propose-only RAG agent; split the monolith (the single all-in-one app) into
per-flow services behind a BFF; lock down secrets with
Secret Manager + WIF.
Stage 4 · Scale & resilience. Only now: go multi-region behind a
global load balancer, move to
AlloyDB (HTAP,
Profile B) for heavy read fan-out + operational
analytics, and add DR (RTO/RPO) + SLOs/observability.
Same Postgres interface (identical SQL), so nothing breaks.