AI INFERENCE
GITOPS.
A cloud-to-homelab delivery pipeline. Built in GitHub. Stored in AWS. Reconciled in the homelab.
- MAX UNAVAILABLE PODS
- 0
- SURGE POD ALLOWED
- 1
GitHub Actions builds the FastAPI inference proxy into private AWS ECR. ArgoCD reconciles its Helm chart into k3s on Proxmox. A rolling-update policy uses maxUnavailable: 0 and maxSurge: 1; readiness gates traffic on Ollama backend availability.
- 01GitHubBUILD / PUSH
- 02AWS ECRPRIVATE REGISTRY
- 03ArgoCDRECONCILE
- 04Proxmox / k3sHYBRID RUNTIME
- 05FastAPI + OllamaINFERENCE / PROBES