Insights, best practices, and lessons learned from building cloud infrastructure at scale
14 posts published
On April 4, 2026, Anthropic silently revoked subscription billing for third-party AI harnesses. Here's the full story of how I rebuilt the request pipeline — from CLI backend to a 7-layer bidirectional proxy — to keep 13 autonomous agents running on my homelab without paying Extra Usage.
Migrating two vLLM models from bare systemd services to a production K3s cluster on the DGX Spark — with NVIDIA GPU Operator time-slicing, ArgoCD app-of-apps GitOps, kube-prometheus-stack monitoring, and Cloudflare Access + Auth0 SSO protecting five web dashboards.
On April 4, 2026, Anthropic silently revoked subscription billing for third-party AI harnesses. Here's the full story of how I rebuilt the request pipeline — from CLI backend to a 7-layer bidirectional proxy — to keep 13 autonomous agents running on my homelab without paying Extra Usage.
Migrating two vLLM models from bare systemd services to a production K3s cluster on the DGX Spark — with NVIDIA GPU Operator time-slicing, ArgoCD app-of-apps GitOps, kube-prometheus-stack monitoring, and Cloudflare Access + Auth0 SSO protecting five web dashboards.
A surgical deep-dive into running an NVIDIA DGX Spark with K3s, multi-agent AI orchestration, three-layer persistent memory (QMD vector search, Graphiti knowledge graph, MuninnDB cognitive memory), and 20 Docker containers on Unraid — all wired together with MCP servers, HashiCorp Vault, Cloudflare Access, and a custom API layer.
Peter Steinberger's OpenClaw isn't just the hottest open-source project of 2026. It's a blueprint for how a single engineer can ship like a team, and why taste and architecture now matter more than raw coding skill.
A deep dive into the sophisticated social engineering attack that targeted security expert Chris Kennedy, and the critical lessons we can all learn about modern scammer tactics, MFA protection, and organizational security awareness.
How an independent audit agent creates separation of powers for AI-driven infrastructure—preventing runaway automation while enabling autonomous operations at scale.
How a single AI agent orchestrates AWS Global WAN infrastructure with autonomous decision-making, separation-of-powers governance, and 10-100x operational acceleration.
How golden paths, AI copilots, and zero-trust guardrails transform platform engineering from gatekeeper to enabler—shipping accountable GenAI workflows at scale.
How the Model Context Protocol turns your private networks into composable context streams that keep AI assistants grounded, auditable, and fast.
Step-by-step topology for exposing n8n through Cloudflare, OPNsense HAProxy, and Unraid with the right headers, WebSocket support, and TLS expectations.
An operator's playbook for shipping responsible GenAI assistants on AWS by mixing Bedrock Guardrails, event-driven monitoring, and zero-trust platform controls.
Part 2 of 2: Extending your Cloud WAN architecture with AWS Network Firewall service insertion for east-west and north-south traffic inspection using Terraform modules and policy-based routing.
Part 1 of 2: A concise, production‑ready README for evolving from a single‑Region deployment to a segmented, multi‑Region AWS Cloud WAN using Terraform modules and Terraform Cloud workspaces.
How I trained a Reachy humanoid robot to see, understand, and interact with the physical world using NVIDIA DGX Spark at home. This isn't just chat anymore—it's embodied AI that truly perceives reality beyond 1s and 0s.