The AWS mirror of the GCP design — same lanes, same contracts, AWS-native services. Profile A · Elastic-Serverless (Lambda/Fargate + Aurora Serverless) and Profile B · High-Scale Transactional & Analytics (ECS/EKS + Aurora provisioned (fixed, always-on database capacity) + SageMaker/Bedrock).
Layer guide (click any term): ingress · ALB · CloudFront/CDN · Aurora · SQS FIFO · EventBridge · Bedrock · OLTP.
| Layer | Choice | Why |
|---|---|---|
| Compute | Lambda + API Gateway / App RunnerECS/EKS on Fargate | lowest operational burden; suits spiky trafficML + long workers |
| OLTP | Aurora Serverless v2 (PostgreSQL)Aurora provisioned (multi-AZ, replicas) | scales to near-zerosustained load |
| RAG vectors | OpenSearch ServerlessAurora pgvector (co-located) | managed scalevectors live beside policy rows — one database for joins + audit |
| Analytics | Redshift / Athena over S3 | actuaries build rates here; results are copied to the live database so the hot path never waits |
The domain model + service contracts are portable (that removes rewrite risk), but identity, networking, eventing semantics (how events are ordered and delivered), the vector backend, secret wiring, deploy mechanism, and observability all change per cloud. We claim portable business logic + per-cloud infra adapters, not lift-and-shift (copying a system unchanged onto another cloud). See the side-by-side.