Skip to content

Customer handoff example

This example uses a fake customer organization, acme-ai, to show the expected handoff flow without assuming a cloud provider.

Inputs

  • Customer Git mirror: https://github.com/acme-ai/private-ai-platform-kit.git
  • Target revision: v0.28.0
  • GPU profile: nvidia
  • Runtime model: Qwen/Qwen3-Coder-Next
  • Secret integration: customer-operated External Secrets or an equivalent mechanism. The checked-in external-secrets.yaml is an example and is not synced by the customer application set.

Configure the overlay

make customer-overlay \
  CUSTOMER_REPO_URL=https://github.com/acme-ai/private-ai-platform-kit.git \
  CUSTOMER_REVISION=v0.28.0 \
  CUSTOMER_GPU_PROFILE=nvidia

The command edits tracked Argo CD manifests. Run it in the customer fork or a deployment branch, then review and commit the diff before syncing.

Review the generated changes in:

  • deploy/gitops/argocd/root-app-customer.yaml
  • deploy/clusters/customer/apps.yaml
  • deploy/clusters/customer/values/inference-gateway.yaml
  • deploy/clusters/customer/values/vllm-nvidia.yaml
  • deploy/clusters/customer/values/rag-service.yaml

Customer decisions

  • Confirm the active kubeconfig points at the intended cluster and that Argo CD, Kyverno, and the required CRDs are already installed.
  • Confirm GPU nodes expose nvidia.com/gpu and have platform.ai/node-pool=gpu.
  • Confirm API-key hashes are sourced from the customer secret backend; do not apply the example secret manifest unchanged.
  • Replace model provenance source-reference digests with customer model-store artifact digests.
  • Replace sample RAG knowledge with approved customer documents and matching Qdrant dimensions.
  • Review agent workspace egress against platform/network/egress-catalog.yaml.

Handoff evidence

After sync, run the in-cluster smoke checks. Point the eval and load test at the customer ingress or an approved port-forward:

make smoke RUNTIME_BACKEND=vllm
make rag-smoke
GATEWAY_URL=https://gateway.example.test PLATFORM_API_KEY=<key> make eval
GATEWAY_URL=https://gateway.example.test PLATFORM_API_KEY=<key> \
  MODEL_ID=Qwen/Qwen3-Coder-Next RUNTIME_BACKEND=vllm make loadtest
make evidence LIVE=1
make release-gate-strict

Replace the example URL, key, and model. Attach the generated Markdown reports and keep the JSON evidence with the release or handoff record.