Technical articles aimed at cloud architects, Kubernetes specialists and engineering leadership. Runnable code, reasoned trade-offs, and what the field teaches that no documentation mentions.
In a textbook RAG the corpus sleeps and indexing runs overnight. As soon as documents and questions arrive as streams, index freshness becomes a latency constraint. Three architectures, their millisecond budget, and the one that holds in production.
Bringing inference and training back behind your firewall takes more than buying graphics cards. Scheduling, GPU sharing, storage, networking, sovereignty: the architecture that works and the traps that cost dearly.
The MCP protocol standardises how a model calls your business tools. What remains is porting it to Kubernetes without turning every MCP server into a back door into your information system.
A manufacturer and its supplier exchange conformity certificates on an immutable ledger. The full modelling, from MSP to chaincode, automated with Hyperledger Bevel 1.3.0 on managed AKS clusters.
Training a model on your cluster's metrics and logs, deploying it, monitoring it and retraining it when it drifts — the full MLOps chain, declarative and reproducible.
Scale to zero, canary rollout, built-in drift capture: KServe brings to inference what Knative brings to applications. Full implementation on a hand-written model.
One autonomous cluster per data centre, federated by a single control plane. How Gardener industrialises the lifecycle of dozens of clusters, and why the stretched cluster remains a bad idea.
Open-source hyperconvergence, virtual machines driven by KubeVirt, RKE2 clusters delivered in fifteen minutes. The full architecture of a cluster factory, from bare metal to self-service catalogue.
The native VPA kills pods to resize them. Ballast smooths the recommendations, Attune applies them in place. Installation, tuning and guardrails of a chain that reclaims 30 to 40% of resources.
A Virtual Kubelet provider that intercepts pods and runs them as lightweight virtual machines under Cloud Hypervisor or Firecracker. Hardware isolation without leaving the Kubernetes API.
An immutable system built, versioned and shipped as an OCI image. Network boot, atomic upgrade with rollback, embedded K3s: the approach that makes edge estates manageable.
Kairos holds the physical host as an immutable system, vk-cocoon isolates every workload in a MicroVM. Two complementary layers for running untrusted code on sites nobody guards.
No article in this category yet.
These articles come out of real engagements. If one of them speaks to your context, let's talk.