Elara
An evolving assistant that connects AI workflows, infrastructure observation, voice and everyday operational tools, with deliberate limits on identity and actions.
Personal project portfolio
IT support by day. After hours, a self-hosted lab of my own: a three-node cluster, an AI assistant called Elara, and a pile of interfaces that make it all easy to read.
Right now
Studying for a B.S. in Data Analytics
Teaching Elara to listen hands-free
Sketching an interactive map of the lab
The lab in numbers
24
containers on 3 Proxmox nodes
+ 3 Raspberry Pis, 1 Gbps in
01 / The lab
Pick a layer to see what lives in it. Addresses, credentials and private services are left out on purpose.
02 / Projects
Each visual is an illustration of the idea, not a screenshot.
An evolving assistant that connects AI workflows, infrastructure observation, voice and everyday operational tools, with deliberate limits on identity and actions.
A three-node environment for running applications and learning how services depend on each other.
Private catalogs, readable product presentation and role-aware portals, described here without exposing private content.
Follows product availability and delivers useful alerts. No retailer partnership or special API access is implied.
Two Raspberry Pi displays that make service state easy to read, one passive, one touch-controlled.
A self-hosted library, playlist mirroring and library upkeep, built around existing open-source services.
Self-hosted media and photo services, with attention to library organization and playback devices.
Mobile folders linked to self-hosted storage, verifying every copy before an original is removed.
Backups, resource-pressure alerts, service checks and staged power management. The goal is understandable recovery: know what is protected, verify what can be restored, and tell a running process from a working service.
03 / Decisions
Every choice trades something. Here is what I picked, why, and what it costs.
A two-node cluster loses quorum when either node drops. A third node removed that fragility.
More hardware, and every node has to stay on matching cluster-stack versions.
OPNsense runs on a dedicated node with an internal Intel NIC, after a USB adapter capped throughput far below the 1 Gbps plan.
The routing node hosts routing and little else, so it is not a free-for-all box.
I kept the ISP gateway in router mode while proving OPNsense, so a rollback stayed one change away. Bridge mode comes once it has proven stable.
An extra layer to reason about until the lab has a single edge.
Pi-hole on its own Pi is the only DNS and DHCP server, so name resolution never depends on a single hypervisor.
It is a single point of failure, so it shuts down last. A stale gateway setting there once broke DNS after a renumber; auditing hardcoded gateways is now part of every change.
When a data disk went read-only, I diagnosed before replacing it. It came back after a rescan and passed SMART and a self-test, which pointed to a cable or connector fault. The data was copied off first.
It also exposed that the service had no backup of its own.
A small agent on each node reports temperature, load and SMART health to the dashboards and to Elara.
Next: long-term metric history does not cover every node yet.
04 / The person behind the projects
Friends online know me as Wren.
I work in IT support by day, and what I care about most is systems people can rely on. These projects are where I practice that on my own terms: solving practical problems and understanding the systems I use.
I like the space between disciplines, where a network makes an application possible, an automation removes a chore, or a better interface makes something complicated feel approachable.
05 / Get in touch
Questions about a project, shared interests, or a good idea. Email me.
sbarba@athelara.com ↗