README.md

AI platform engineering documentation

The optional AI platform adds an Ollama inference runtime, LLM gateway, agent orchestrator, Redis rate limiting, frontend assistant integration, observability, Istio policy, and an optional Argo Rollouts canary to Online Boutique.

Environment support: the end-to-end AI bootstrap and GitOps path is currently production-only. The AI workflows accept only prod, the optional Argo CD Application tracks main, and AI resources are not part of the dev root app. A dev Terraform node-pool definition exists, but it is not a supported complete dev AI deployment workflow.

Implementation

Layer	Repository source
Infrastructure	terraform/live/prod/ai-node-pool.tf and ai-iam.tf
Runtime and services	kustomize/ai-platform/base
CPU/model profiles	kustomize/ai-platform/models/cpu-small and cpu-better
Security	kustomize/ai-platform/security
Canary delivery	kustomize/ai-platform/rollout
GitOps entrypoint	argocd/optional-apps/ai-platform/manifests/ai-platform.yaml
Automation	.github/workflows/ai-platform-bootstrap.yml, ai-nodepool-disable.yml, and ai-platform-ci.yml

The base includes namespaces, RBAC, a persistent model cache, Ollama and model bootstrap, the gateway and orchestrator, rate-limit Redis, ServiceMonitors, Prometheus rules, a Grafana dashboard ConfigMap, and the security layer. The rollout overlay removes the normal gateway Deployment and substitutes a Rollout and Prometheus AnalysisTemplate.

Request flow

Frontend assistant
        |
        v
AI agent orchestrator ----> Kubernetes API / Prometheus / Elasticsearch
        |
        v
LLM gateway ----> rate-limit Redis
        |
        v
Ollama model runtime

Documentation map

• Architecture
• Components
• System design
• Workflow
• Deployment guide
• Cost control
• AI observability
• Operations runbook
• Troubleshooting
• Disaster recovery
• Demo scenarios
• Lessons learned
• Project showcase

Safe local validation

Rendering does not connect to a cluster:

kubectl kustomize kustomize/ai-platform/models/cpu-small >/tmp/ai-small.yaml
kubectl kustomize kustomize/ai-platform/models/cpu-better >/tmp/ai-better.yaml
kubectl kustomize kustomize/ai-platform/rollout >/tmp/ai-rollout.yaml

Deploy only through the documented production bootstrap after reviewing GCP costs, node-pool capacity, required secrets, and workflow approvals.