Phase IV – Production Cloud Migration & GitOps

Deploy Edge Configurations to Cloud Prod (e.g. AWS EKS)

To run this in production, use Terraform to provision an AWS EKS cluster with auto-scaling GPU node groups (g5 or p4d instances) and an S3 bucket for storing model weights.

Organize your Kubernetes setup using Kustomize. Keep core deployment logic in deploy/base/ and cloud-specific changes in deploy/overlays/aws-eks/. Replace local hostPath mounts with persistent volume claims that use the AWS S3 CSI driver to load weights directly from your bucket.

Git Push to Repository
       │
       ▼
ArgoCD Sync Engine
       │
       ▼
AWS EKS Cluster
   ├── AWS ALB Ingress Gateway
   ├── Auto-Scaling Multi-GPU Nodes
   └── S3 CSI Driver ◄── Model Weights (s3://...)

Finally, set up ArgoCD to watch your Git repository. Whenever you commit changes to configurations, schemas, or deployment manifests, ArgoCD automatically syncs them to your EKS cluster without manual intervention.

More

• Detailed Technical Steps
• Cloud-Agnostic Multi-Modal Model Serving Engine
• Phase I: Host Prerequisites & K3s Runtime Patching
• Phase II – Declarative Model Configuration
• Phase III – Containerization & Local Deployment