Uses
The hardware, software, and habits I actually use. Format borrowed fromuses.tech.
Homelab hardware
The hypervisor. VMs, LXCs, and the K3s nodes all live here.
A small Kubernetes cluster for trying things I do not want to try at work.
Disks, backups, and the family photo pile.
How I reach the lab from anywhere. MagicDNS and subnet routes do most of the work.
Network gear
Separate VLANs for the lab, IoT junk, and the stuff I actually care about.
Software & tooling
Where I write scripts, infra, and this site.
CI at work, and the pipelines for the lab.
What I reach for first. Security automation, AI glue, and anything boring and repetitive.
Lights, sensors, and a pile of automations.
Photo library I host for family and friends.
AI & ML
Trying models and wiring them into real workflows, not just demos.
Retrieval Architectures
Getting the right context in front of a model. Reranking, query tweaks, and retrieval that is not always a vector database.
MLOps Pipelines
Eval, versioning, deploys, and a way to roll back when a prompt change makes things worse.
Self-hosted Inference
Local models on the homelab GPU, mostly so I can experiment without sending everything out.
Cloud & DevOps stack
The cluster in the rack. Apps ship as Helm charts.
Git says what should be running. Flux makes the cluster match.
Building images and running things locally.
I ran this stack for about four years. I still help with it.
PostgreSQL + Redis
Databases and cache for stuff on the homelab.
Workflow
GitOps with FluxCD
A merge deploys it. I try not to kubectl apply in production.
Validation gates in CI
Changes get checked, reviewed, and a rollback path before they go out.
Segregated dev and prod
Dev and prod are separate, and the prod config is encrypted.
Runbooks for everything
If I do it by hand twice, I write it down.