Real World Example

• Startups often put the LLM backend on Cloud Run or a container platform and the frontend on Vercel or Firebase Hosting.
• Banks and large enterprises in India often run models through cloud regions in India or on private infrastructure for data rules.
• Every serious team has a staging environment with its own keys and an eval run before production deploys.

git push main
   │
   ▼
GitHub Actions: lint + unit tests + 60-case eval (must pass)
   │
   ▼
build image ──► Artifact Registry ──► Cloud Run (asia-south1), 10% traffic
   │
   ▼
watch p95, errors, thumbs down for 30 min ──► 100% traffic  (or roll back)

# .github/workflows/deploy.yml (deploy job)
deploy:
  needs: [tests, evals]
  runs-on: ubuntu-latest
  permissions: { id-token: write, contents: read }
  steps:
    - uses: actions/checkout@v4
    - uses: google-github-actions/auth@v2
      with:
        workload_identity_provider: ${{ secrets.GCP_WIF_PROVIDER }}
        service_account: [email protected]
    - uses: google-github-actions/setup-gcloud@v2
    - run: |
        gcloud run deploy shopease-assistant --source . --region asia-south1 \
          --set-secrets ANTHROPIC_API_KEY=anthropic-key:latest --no-traffic --tag canary
        gcloud run services update-traffic shopease-assistant --region asia-south1 --to-tags canary=10