BuiltForProd Blueprints
A BuiltForProd Blueprint is an opinionated, production-ready reference architecture for a real-world workload, delivered as repositories and deployed into the workload stages of the Baseline. The Blueprint gives you the production architecture for the use case; your team builds the application. Each Baseline edition has three, Web App, Data and ETL, and Secrets, and each is a working system, not a diagram.
How a blueprint sits on the Baseline
A blueprint never creates what the landing zone owns. It reads the network, DNS and registry values the Baseline publishes per stage, the landing-zone contract, and it never reads the Baseline's state. Each blueprint repository has one CI identity per stage in the auto account, subscription or project, which accepts only jobs running in that stage's GitHub Environment and reaches only that stage; a code repository adds a build identity that can only push images. Each stage of a blueprint has its own state prefix in the Baseline's state store, and the blueprint has its own plan and apply workflows. Blueprint repositories can therefore be planned and applied independently of the landing zone.
| Edition | Landing-zone contract | CI identities | Registry | SOPS keys |
|---|---|---|---|---|
| AWS | SSM Parameter Store under /acme/usw2/<stage>/...; blueprints publish their own outputs the same way | A role per blueprint repository and stage in core-auto, such as acme-aws-blueprint-webapp-infra-dev-deployer, assumed through GitHub OIDC; it may assume only that stage account's deployer | ECR in core-artifacts | KMS keys per stage in core-security |
| Azure | The contract Key Vault kv-acme-eus2-<stage>-plat, 30 secrets per stage | A user-assigned managed identity per repository and stage in acme-core-auto, such as id-acme-azure-blueprint-webapp-infra-dev-deployer, federated to that stage's GitHub Environment; the infrastructure identities hold the role acme-blueprint-deployer on their stage's plat subscription only | Azure Container Registry in acme-core-artifacts | Key Vault keys per stage in acme-core-security |
| GCP | Parameter Manager parameters in each plat project, 24 per stage | A CI service account per repository and stage in acme-core-auto, such as sa-acme-webapp-infra-dev-ci, reached through Workload Identity Federation from that stage's GitHub Environment; the infrastructure accounts impersonate the deployer of their stage's plat project only, the code accounts hold direct grants on what they release to | Artifact Registry in acme-core-artifacts | Cloud KMS keys per stage in acme-core-security |
The AWS blueprints
The blueprint catalog lists every blueprint with the cloud it runs on. These are the AWS blueprints in detail.
| Blueprint | Repositories | What it deploys per stage |
|---|---|---|
| Web App Blueprint | acme-aws-blueprint-webapp-infra, acme-aws-blueprint-webapp-code, acme-aws-blueprint-webapp-gitops | An EKS cluster with a managed node group, Karpenter and Cluster Autoscaler; DocumentDB and ElastiCache Redis; the AWS Load Balancer Controller, ExternalDNS (public and internal), External Secrets Operator, Fluent Bit and ArgoCD; ACM certificates; a private S3 bucket behind CloudFront for the React front end. The code repository holds a Flask API, its Dockerfile and the build-once, promote-by-tag pipelines; the GitOps repository holds the Helm chart and per-stage values ArgoCD syncs |
| Data and ETL Blueprint | acme-aws-blueprint-etl-infra, acme-aws-blueprint-etl-code | A zoned S3 data lake (raw, processed, curated) with SSE-KMS and versioning; Glue Data Catalog databases per zone; a PySpark Glue job and crawler; Lake Formation governance with LF-Tags; a VPC-attached Lambda trigger that starts the job when a file lands in the raw bucket. The code repository holds the Glue script, the Lambda container and its pipelines |
| Secrets Blueprint | acme-aws-blueprint-secrets | One SOPS-encrypted YAML per application per stage, encrypted with that stage's KMS key from the security account; a validation workflow on pull requests and a sync workflow that writes SSM Parameter Store SecureStrings in the target account on merge. Access is enforced by KMS key policy: leads can use every stage key, engineers only sandbox and dev |
| Relational App Blueprint | acme-aws-blueprint-relapp-infra, acme-aws-blueprint-relapp-code | A containerized API on ECS Fargate with Aurora PostgreSQL Serverless v2 and IAM database authentication, and a React front end behind CloudFront and WAF |
| Multi-tenant SaaS Blueprint | acme-aws-blueprint-saas-infra, acme-aws-blueprint-saas-code | A multi-tenant web app with a host name per tenant behind CloudFront and WAF: a tenant API on ECS Fargate over Aurora PostgreSQL Serverless v2 with row-level security, Amazon Cognito sign-in by invitation or per-tenant SSO, a Step Functions control plane that onboards and offboards tenants, per-tenant usage and cost, and dedicated stacks for premium tenants |
| Serverless and Integration Blueprint | acme-aws-blueprint-serverless-infra, acme-aws-blueprint-serverless-code | An event-driven order service built only from pay-per-request services: API Gateway with WAF, Lambda, an EventBridge bus with replay, a Step Functions workflow, SQS queues with dead-letter alarms and DynamoDB |
| Warehouse and BI Blueprint | acme-aws-blueprint-analytics-infra, acme-aws-blueprint-analytics-code | A private Redshift Serverless warehouse over a Lake Formation governed catalog, with SQL marts, an Athena workgroup and a Quick Sight dashboard deployed by the pipeline |
| GenAI RAG Assistant Blueprint | acme-aws-blueprint-rag-infra, acme-aws-blueprint-rag-code | A chat assistant that answers from your documents with citations: a Bedrock knowledge base on Aurora pgvector, a guardrail, Cognito sign-in with group-based document access, and an evaluation gate |
| Agentic AI Platform Blueprint | acme-aws-blueprint-agents-infra, acme-aws-blueprint-agents-code | A managed agent runtime with a sample operations assistant: AgentCore Runtime, an MCP tool gateway with least-privilege tools, propose-then-confirm for writes, conversation memory, a guardrail and an evaluation gate |
| VM Migration Landing Blueprint | acme-aws-blueprint-vmlanding-infra | A place to land migrated servers: AWS Transform MGN replication into private subnets with the blueprint's instance profile, security group and tags, golden Amazon Linux 2023 and Windows Server 2025 images from EC2 Image Builder with a sample fleet behind an internal Application Load Balancer, patching, agents and backups from day one, and optional disaster recovery replication |
| Backup and DR Blueprint | acme-aws-blueprint-backupdr-infra | AWS Backup plans that protect the stores of every stage by the tags they already carry, copies in the isolated core-backup account and a second Region under vault locks, and a monthly tested restore with alerts |
| FinOps Cost Analytics Blueprint | acme-aws-blueprint-finops-infra | Monthly budgets and cost anomaly alerts in every stage, and Athena cost views by repository, team, stage and service over the organization cost exports, with AI cost views and a tag-coverage report |
| Observability Blueprint | acme-aws-blueprint-observability-infra | Service level objectives with two-window burn-rate alerts, health checks of the published URLs, CloudWatch dashboards and OpenTelemetry endpoints for each stage, alerting the stage operations topic |
The Web App Blueprint's sixteen units are grouped by the webapp-stage template and the ETL Blueprint's six by etl-stage; each stage file instantiates its template with only the values that differ per stage. Staging and prod run the high-availability layout (multi-AZ data stores, ArgoCD HA, three-node minimum), dev the minimal one.
The Azure blueprints
| Blueprint | Repositories | What it deploys per stage |
|---|---|---|
| Web App Blueprint | acme-azure-blueprint-webapp-infra, acme-azure-blueprint-webapp-code, acme-azure-blueprint-webapp-gitops | An AKS cluster with a private API server, Azure CNI Overlay with Cilium, Microsoft Entra ID and Azure RBAC for Kubernetes, and Node Auto Provisioning for application nodes; the managed NGINX ingress; Front Door with a web application firewall as the only public entry; Cosmos DB for MongoDB vCore and Azure Managed Redis behind private endpoints; an application Key Vault; Application Insights; External Secrets, cert-manager and ArgoCD. The code repository holds the Flask API and the React front end with their pipelines; the GitOps repository holds the Helm chart and per-stage values ArgoCD syncs |
| Data and ETL Blueprint | acme-azure-blueprint-etl-infra, acme-azure-blueprint-etl-code | An Azure Data Lake Storage Gen2 lake with raw, processed, curated and managed zones behind private endpoints; an Azure Databricks Premium workspace injected into the stage network; Unity Catalog with a catalog per stage; the Databricks job; and the trigger: an Event Grid system topic, a Storage queue and an event-driven Container Apps job. The code repository holds the PySpark transform and the trigger container with their pipelines |
| Secrets Blueprint | acme-azure-blueprint-secrets | One SOPS-encrypted YAML per application per stage, encrypted with that stage's Key Vault key; a validation workflow on pull requests and a sync workflow that writes each key as a secret in the stage's application Key Vault on merge. Access is enforced by Key Vault RBAC: every engineer group can decrypt sandbox and dev, only the platform leads and the DevOps leads staging and prod |
| Relational App Blueprint | acme-azure-blueprint-relapp-infra, acme-azure-blueprint-relapp-code | A containerized API on App Service with slot-swap releases, Azure SQL Database with Entra-only access and a single-page front end behind Front Door |
| Multi-tenant SaaS Blueprint | acme-azure-blueprint-saas-infra, acme-azure-blueprint-saas-code | A multi-tenant web app on App Service with a host name per tenant behind Front Door: row-level security in Azure SQL Database, Microsoft Entra External ID sign-in with membership by invitation, a control plane that onboards, meters and offboards tenants, and dedicated stacks for premium tenants |
| Serverless and Integration Blueprint | acme-azure-blueprint-serverless-infra, acme-azure-blueprint-serverless-code | An orders API behind API Management on Azure Functions Flex Consumption, with a Durable Functions workflow, Service Bus queues with dead-letter replay, an Event Grid topic and serverless Cosmos DB, private in staging and prod |
| Warehouse and BI Blueprint | acme-azure-blueprint-analytics-infra, acme-azure-blueprint-analytics-code | A Microsoft Fabric capacity, lakehouse and warehouse per stage, with SQL marts and a Power BI report deployed by the pipeline |
| GenAI RAG Assistant Blueprint | acme-azure-blueprint-rag-infra, acme-azure-blueprint-rag-code | A chat assistant behind Microsoft Entra sign-in that answers from your documents with citations, using Azure AI Search, Azure OpenAI models and an evaluation gate |
| Agentic AI Platform Blueprint | acme-azure-blueprint-agents-infra, acme-azure-blueprint-agents-code | An operations assistant on Foundry Agent Service that answers from its MCP tools and runs every write only after the signed-in user confirms it, with an evaluation gate |
| VM Migration Landing Blueprint | acme-azure-blueprint-vmlanding-infra | An Azure Migrate project and a rehost resource group where migrated servers land on Azure VMs, golden Ubuntu and Windows Server images built by VM Image Builder with a sample fleet, and Azure Update Manager patch windows, monitoring agents and backups from day one |
| Backup and DR Blueprint | acme-azure-blueprint-backupdr-infra | An Azure Backup vault in every stage with daily vaulted backups and backup audits, locked, geo-redundant and behind multi-user authorization in staging and prod, where a monthly tested restore alerts on failure |
| FinOps Cost Analytics Blueprint | acme-azure-blueprint-finops-infra | Monthly budgets and cost anomaly alerts in every stage, tag inheritance and saved cost views, and FOCUS cost exports read by Power BI reports and a monthly review script |
| Observability Blueprint | acme-azure-blueprint-observability-infra | Service level objectives with two-window burn-rate alerts, availability tests from three locations, Azure Monitor workbooks, a Prometheus workspace and an OpenTelemetry endpoint for each stage, alerting the stage action group |
| Azure Virtual Desktop Blueprint | acme-azure-blueprint-avd-infra | A pooled Azure Virtual Desktop host pool whose session hosts the service builds, updates and scales itself, a workspace with one full Windows 11 desktop, FSLogix profiles on Azure Files over Microsoft Entra Kerberos and Azure Virtual Desktop Insights |
The Web App Blueprint's thirteen units are grouped by the webapp-stage template and the ETL Blueprint's five by etl-stage. Staging and prod run the high-availability layout (zone-redundant Cosmos DB and Managed Redis, ArgoCD HA, the Standard AKS tier), dev the minimal one.
The GCP blueprints
| Blueprint | Repositories | What it deploys per stage |
|---|---|---|
| Web App Blueprint | acme-gcp-blueprint-webapp-infra, acme-gcp-blueprint-webapp-code, acme-gcp-blueprint-webapp-gitops | A regional GKE Standard cluster with private nodes, Dataplane V2, Workload Identity Federation for GKE and node auto-provisioning; the API edge the chart's GKE Gateway uses (a global address, a Google-managed certificate and a Cloud Armor policy); the front end on Cloud Run behind a global external Application Load Balancer with Cloud CDN and Cloud Armor; Firestore with MongoDB compatibility and Memorystore for Valkey without passwords; External Secrets, external-dns, cert-manager and ArgoCD. The code repository holds the Flask API and the React front end with their pipelines; the GitOps repository holds the Helm chart and per-stage values ArgoCD syncs |
| Data and ETL Blueprint | acme-gcp-blueprint-etl-infra, acme-gcp-blueprint-etl-code | Raw, processed and curated Cloud Storage buckets; a Dataplex Universal Catalog lake whose discovery publishes the files as BigQuery tables; BigQuery datasets with a policy-tag taxonomy for column-level access; the Spark service account; and the trigger: Eventarc on the raw bucket and an internal Cloud Run service that starts one Dataproc Serverless batch per input file. The code repository holds the PySpark transform and the trigger service with their pipelines |
| Secrets Blueprint | acme-gcp-blueprint-secrets | One SOPS-encrypted YAML per application per stage, encrypted with that stage's Cloud KMS key; a validation workflow on pull requests and a sync workflow that writes each key as a Secret Manager secret in the stage project on merge. Access is enforced by Cloud KMS IAM: every engineer group can decrypt sandbox and dev, only the platform leads and the DevOps leads staging and prod |
| Relational App Blueprint | acme-gcp-blueprint-relapp-infra, acme-gcp-blueprint-relapp-code | A Flask API and a React front end on Cloud Run behind one global external Application Load Balancer with Cloud Armor, a Google-managed certificate and Cloud CDN; Cloud SQL for PostgreSQL 18 reached only over Private Service Connect with IAM database authentication; Alembic migrations run as a Cloud Run job before every release. The code repository holds the API, the migrations and the front end with their pipelines |
| Multi-tenant SaaS Blueprint | acme-gcp-blueprint-saas-infra, acme-gcp-blueprint-saas-code | A multi-tenant web app on Cloud Run and Cloud SQL for PostgreSQL with a host name per tenant behind a global load balancer with Cloud Armor: row-level security, invite-only Identity Platform sign-in with per-tenant SSO, a control plane on Workflows that onboards and offboards tenants, per-tenant usage metering and dedicated stacks for premium tenants |
| Serverless and Integration Blueprint | acme-gcp-blueprint-serverless-infra, acme-gcp-blueprint-serverless-code | An event-driven orders platform on Cloud Run behind a global external Application Load Balancer with Cloud Armor: Pub/Sub events, a Workflows saga started by Eventarc that compensates on failure, queues with dead letters and replay, and Firestore |
| Warehouse and BI Blueprint | acme-gcp-blueprint-analytics-infra, acme-gcp-blueprint-analytics-code | BigQuery staging, marts and assertion datasets built by a Dataform repository from the code repository at a pinned commit, with scheduled refreshes; a sample source in standalone mode or the Data and ETL Blueprint table through an optional switch; the runner and BI reader identities; a Looker Studio report created from the published Linking API URL; an optional BigQuery reservation |
| GenAI RAG Assistant Blueprint | acme-gcp-blueprint-rag-infra, acme-gcp-blueprint-rag-code | A chat assistant that answers from your documents with citations: a Cloud Storage document bucket whose new objects are chunked and embedded by a Pub/Sub-triggered Cloud Run job; Cloud SQL for PostgreSQL 18 with pgvector over Private Service Connect, where retrieval filters each document by its Google groups; Gemini models on Vertex AI behind a Model Armor template that fails closed; a Cloud Run chat API and UI behind a global load balancer with Identity-Aware Proxy and Cloud Armor; a BigQuery prompt log. The code repository holds the API, the ingestion, the evaluation harness with its golden set and the pipelines that gate every release on it |
| Agentic AI Platform Blueprint | acme-gcp-blueprint-agents-infra, acme-gcp-blueprint-agents-code | An operations assistant built with the Agent Development Kit on Vertex AI Agent Engine, with Sessions, Memory Bank and a Private Service Connect interface; an MCP tool server on Cloud Run whose write tool only proposes, and an internal action runner that runs a proposal once after the signed-in user confirms it; two Firestore databases with per-tool IAM; Gemini behind a Model Armor template that fails closed; a Cloud Run chat UI behind Identity-Aware Proxy and Cloud Armor; a BigQuery prompt log; Cloud Trace. The code repository holds the agent, the tools, the UI, the evaluation scenarios and the gated pipelines |
| VM Migration Landing Blueprint | acme-gcp-blueprint-vmlanding-infra | Migrate to Virtual Machines connected to the landing zone, so each migrated server comes up on the private vms subnet with its own identity, Shielded VM, OS Login and no public address; golden images with a sample fleet behind an internal load balancer; VM Manager patching, the Ops Agent, IAP access and backups to the isolated vault from day one |
| Backup and DR Blueprint | acme-gcp-blueprint-backupdr-infra | Cloud SQL backups in the isolated Backup and DR vault and Firestore backup schedules in every stage, with a daily backup check and a monthly tested restore with alerts |
| FinOps Cost Analytics Blueprint | acme-gcp-blueprint-finops-infra | A daily cost anomaly alert next to the Baseline budgets, and BigQuery cost views by repository, team, stage and service over the Cloud Billing export, with FOCUS columns, AI cost views and a Looker Studio report |
| Observability Blueprint | acme-gcp-blueprint-observability-infra | Service level objectives with two-window burn-rate alerts, uptime checks of the published URLs, log alerts, Cloud Monitoring dashboards and an OpenTelemetry endpoint for each stage, alerting the stage operations mailbox |
The Web App Blueprint's eleven units are grouped by the webapp-stage template and the ETL Blueprint's five by etl-stage. Staging and prod spread the cluster over three zones, give the Valkey cache a replica in a second zone and run ArgoCD HA; prod adds daily database backups; dev runs the smallest layout.
Delivery inside a blueprint
Application images are built once on merge to main, pushed to the registry in the artifacts account, subscription or project (ECR, Azure Container Registry, Artifact Registry) under an immutable main-<short sha> tag, given a vX.Y.Z tag at release, and rolled out by pull request against the GitOps repository: dev and staging merge themselves and ArgoCD auto-syncs; prod is merged by a person and synced manually. On GCP the front end is promoted as a Cloud Run revision. The ETL release pipeline records the trigger image tag it deployed in the contract store (an SSM parameter, a contract Key Vault secret or a Parameter Manager parameter), and the infrastructure reads it back. The GitOps and immutable artifacts pages describe the mechanism.
Using a blueprint as-is
Full Deployment with Blueprints includes the Baseline and any two blueprints deployed by the BuiltForProd team; Blueprint Deployment adds one or more blueprints to an existing Baseline. The blueprint catalog lists every blueprint to choose from. What you receive is a working system in dev, staging and prod: the web application answers on its public domain behind a web application firewall, the data pipeline processes a file dropped into the raw zone, the secrets sync writes values an application can read. Your teams use it as the reference for their own workloads: the same repository layout, the same pipelines, the same conventions.
Customizing a blueprint
Blueprints are templates, so every deployment-specific value is a tagged setting rather than an assumption in code:
git grep "TODO: "lists what must be set. On AWS: domains, ECR registry, GitHub variables and secrets, the stage VPC CIDRs mirrored in the chart's NetworkPolicy. On Azure, for exampleaks_admin_group_object_ids, the Entra groups that administer each cluster. On GCP, for example the identities allowed to upload input files to the raw bucket.git grep "@optional: "lists what may change. On AWS: node counts, DocumentDB instance count and deletion protection, Redis nodes and multi-AZ, ArgoCD HA and auto-sync, log retention, Glue worker type and count, Lambda memory and timeout, HPA and PDB settings. On Azure, for examplefront_door_skuandmongo_compute_tier. On GCP, for examplegke_zones,firestore_backup_retention_daysandvalkey_replicas.- Adding a resource follows the module extensibility guide: a module named by resource type that accepts the standard inputs, a unit definition with its dependencies mocked, a
unitblock in the stage template, and a per-stage value in each stage file. On AWS, security groups are separate units injected into the resources that use them. - Creating a new blueprint copies the structure of an existing one (the ETL blueprint is the smallest) and registers it in the landing zone first. On AWS: its per-stage roles in the
github-oidcunit, ECR repositories and Identity Center groups. On Azure: its repository in thegithub-oidcunit, which creates the per-stage identities, with the registry and vault grants made by the units that own them. On GCP: its repository and stages in theworkload-identityrepositories, which create the per-stage CI accounts, and its Artifact Registry repository; the web app also needs its stages inwebapp_gke_stagesofplat.hcl.
The Baseline page lists the deployment options; BuiltForProd Managed is the option when you want BuiltForProd engineers to do the customizing.