PinnedRupinder Guron·Oct 12A Real Two-Model Canary Rollout for RAG Reranking, with KServeMost “canary deployment” posts for ML models show the same model deployed twice behind a traffic split. That proves your routing works. It…
PinnedRupinder Guron·Aug 19AI-Native Platform Engine: Local LLM Inference Control PlaneAI Native Platform Engine
Rupinder Guron·Oct 12Keyless CI/CD: Authenticating GitHub Actions to GKE Without a Stored Service Account KeyThe default way to let a CI pipeline talk to GCP is to generate a service account JSON key, paste it into a repo secret, and have the…
Rupinder Guron·Oct 12GKE Autopilot on a Free-Tier Budget: A Real Capacity Ceiling, DiagnosedMost posts comparing GKE Autopilot and Standard mode are abstract — a feature table, a paragraph on “control vs. convenience.” This isn’t…