Iulian Flavius M.

Most servers can't be rebuilt — configured by hand, documented nowhere, now irreplaceable. I fix that. NixOS, Terraform, Incus, Proxmox — declarative by default: no clickops, no snowflake servers, no restore gambles. EU-based, working EU hours. 100% Job Success Score. 5.0 across 67 reviews. My longest engagement is a multi-year platform role with a Nasdaq-listed Swedish fintech, ongoing. My recent Upwork contracts read the same way: NixOS deployments, Proxmox with GPU passthrough, Nextcloud and Collabora, MinIO storage, VPS hardening. THE PLATFORM LAYER NixOS for declarative systems. Terraform + Incus for provisioning. Proxmox and QEMU for virtualization, including VFIO GPU passthrough; I run a multi-GPU inference host myself — NVIDIA and AMD/ROCm. WireGuard and Tailscale for networking. nftables firewalls at the edge. Backups get tested — I don't call a backup working until I've restored from it. WHAT RUNS ON TOP Keycloak and Authentik for single sign-on (SSO), MFA, and identity federation. Nextcloud with Collabora, RustDesk, MinIO, Coolify, and local LLMs — Ollama, llama.cpp, vLLM. Plus TypeScript when the platform needs a face. WHY IT MATTERS Most sysadmins configure by hand: servers end up hand-built, undocumented, and impossible to reproduce. I make infrastructure reproducible: versioned, reviewable, restorable. Security is structural, not a plugin: keys instead of passwords, MFA at the identity layer, audit logging on anything internet-facing. Lately it's also this: AI-built products you now need to own, audit, or escape from — repo, data, and a plan that works without the platform. If you're running servers nobody can rebuild, or moving off SaaS over costs and data control, that's my exact kind of project. When did you last verify a restore actually worked?