Skip to content
AI CROPS Cloud

Resources

Reference architectures

The patterns we start from for platforms, data, AI and operations, with the design decisions that make them hold up in production. They are adapted for each client and are not descriptions of any client's system.

Cloud-native product platform

Cloud-native product platform

How code moves from a commit to running in production on a governed, multi-environment Kubernetes platform.

Reference architecture — not a client deployment
  1. Stage 1: Build

    • Git repository
    • CI pipeline & tests
    • Image & artifact registry
    • Policy & security checks
  2. Stage 2: Deliver

    • GitOps controller
    • Dev → staging → prod promotion
    • Progressive rollout
  3. Stage 3: Run

    • Kubernetes (EKS / GKE)
    • Managed databases
    • Queues & caches
    • API gateway
  4. Stage 4: Observe

    • Metrics, logs & traces
    • SLO alerting
    • Cost per service

Foundation

  • Landing zone: accounts / projects
  • Network
  • IAM & guardrails
  • Central logging
  • Infrastructure as code

Key design decisions

  • Every environment is created from the same Terraform modules; differences are variables, not hand edits.
  • The GitOps controller is the only path to production, so every change has an author, a review and a rollback.
  • Cost and SLO dashboards are per service, so the team that owns a service also sees its cost and reliability.

Modern data platform

Modern data platform

From operational sources to governed analytics, BI and AI-ready data, with quality checks at every layer.

Reference architecture — not a client deployment
  1. Stage 1: Sources

    • Operational databases
    • SaaS applications
    • Event streams
    • Files & documents
  2. Stage 2: Ingest

    • Change data capture
    • Batch connectors
    • Stream ingestion
  3. Stage 3: Store & transform

    • Raw landing zone
    • Lakehouse tables
    • Warehouse
    • dbt models
  4. Stage 4: Consume

    • BI dashboards
    • Semantic layer
    • ML features
    • Data APIs

Foundation

  • Orchestration
  • Data quality tests
  • Catalog & lineage
  • Access control & PII policy

Key design decisions

  • Raw data is kept immutable, so any model can be rebuilt and any number traced back to its source.
  • Metric definitions live in one version-controlled semantic layer shared by every dashboard.
  • The same governed tables feed BI and machine learning, so AI and reporting do not disagree.

Production generative AI application

Production generative AI application

Retrieval-augmented generation and agentic workflows with access control, evaluation and cost visibility built in.

Reference architecture — not a client deployment
  1. Stage 1: Interfaces

    • Product UI
    • Internal tools
    • Public API
  2. Stage 2: Application

    • AI gateway: auth, limits, routing
    • Orchestrator / agent runtime
    • Tool connectors with approval gates
  3. Stage 3: Knowledge & models

    • Permission-aware retrieval
    • Document ingestion pipeline
    • Foundation models
    • Custom / fine-tuned models
  4. Stage 4: Quality & control

    • Evaluation suite in CI
    • Guardrails & PII filters
    • Traces, cost & feedback

Foundation

  • Cloud platform
  • Identity & tenant isolation
  • Data governance
  • Audit log

Key design decisions

  • Retrieval applies the user's own permissions, so the assistant cannot reveal documents the user could not open.
  • Agent tools are allow-listed; sensitive actions pause for human approval and every step is logged.
  • Every prompt, retrieval or model change is scored against the evaluation set before it ships.

Intelligent operations & governance

Intelligent operations & governance

Signals from infrastructure, applications, security and AI usage turned into decisions, automated actions and audit evidence.

Reference architecture — not a client deployment
  1. Stage 1: Signals

    • Metrics, logs & traces
    • Cloud config & audit logs
    • Security findings
    • AI usage events
  2. Stage 2: Process

    • OpenTelemetry collectors
    • Log analytics / SIEM
    • Policy engine
  3. Stage 3: Decide

    • SLO & alert rules
    • AI-assisted triage
    • Approval workflow
  4. Stage 4: Act

    • Runbook automation
    • Tickets & ChatOps
    • Compliance evidence store
    • Reports

Foundation

  • Identity & access
  • Encryption & key management
  • Immutable audit trail

Key design decisions

  • Alerts are tied to SLOs, so on-call engineers are paged for user impact, not for every metric spike.
  • AI-assisted triage summarizes and suggests; only pre-approved, reversible actions run without a human.
  • Control checks run continuously and store their evidence, so audits draw on records instead of screenshots.

Want to see how a pattern would fit your system?

We can walk through any of these architectures against your current setup and constraints.