Panu GuysonDevOps engineer and AI infrastructure architect

I make AI run in production.

Inference, GPU serving, gateways and observability.

From bare-metal servers to Kubernetes across multi-cloud and multi-IDC setups, and now LLM serving on a DGX Spark in my homelab. I build the gateway, GPU and observability layers that make AI run.

AI inferenceGPU servingGatewaysObservabilityHomelab

View work →Read the blog

DGX SPARK ↑ vLLM per container · :8001 :8002 :8003 HOME-BUILT GATEWAY Proxmox VM PROMETHEUS + Grafana CLIENTS → text, code,images
fig. my homelab LLM stack Clients (text, code, images) → home-built gateway on a Proxmox VM → vLLM per container on the DGX Spark (:8001 :8002 :8003) · Prometheus + Grafana for the GPUs
U2

Archive

posts per year

“Engineering is the art of organizing complexity into invisible simplicity.”

philosophy
U3

Work

Selected work, most of it from the homelab.
01 · AI infra · Written up in 2 posts

LLM serving on DGX Spark

Gateway on a Proxmox VM, vLLM per model in its own container, Prometheus and Grafana for GPU metrics.

Read EP.1 and EP.2 →
02 · Energy and IoT

Panya OS

A custom smart home energy management system built for optimized power use and IoT automation.

Active development
03 · Platform

Self-hosted infrastructure

A private cloud of containerized services, media servers and dev tools on self-managed hardware.

04 · IoT

Smart home ecosystem

Home automation that combines IoT sensors, custom scripts and centralized control.

05 · AI infra

Self-hosted AI gateway (unreleased)

A gateway I build myself to route LLM traffic, with per-provider failover, cooldowns, an MCP proxy, and an admin UI.

06 · Delivery

AI-assisted SDLC (in progress)

A delivery workflow where AI agents open PRs, run CI, and review code, with human approval gates before every merge and release.