About
I'm Zoltan. In late 2025, I started building with AI instead of just writing about it. Two months later I had a dual-3090 AI server in my office and an RTX 4080 generating images in 7 seconds. This site is where I document all of it.
Every post covers something I actually built, what went wrong, and what worked. The dead ends are in there too — that's where most of the useful information is.
The goal: when someone searches for a problem I've already solved, they find the answer here instead of piecing it together from five different forum threads. If you're building a homelab, running local AI, or trying to figure out why ComfyUI keeps eating all your VRAM — you're in the right place.
The name: Make and Modify. Build something, find out what breaks, fix it, and document the whole process. That's the loop.
Background
8 years as a Technical Writer at United Airlines. 6.5 years as a Senior Technical Trainer at PROS. I've spent 14 years explaining complex software to people who didn't ask for it — the trick is figuring out what actually matters, ignoring the rest, and writing it so people don't zone out by paragraph two. Now I do the same thing for my own builds. If I spent six hours troubleshooting a VRAM issue, you get the 10-minute version with the actual fix.
Why Local?
Because I don't want my data on someone else's server, I don't want to pay monthly for API credits, and I wanted to see if consumer GPUs could actually run the models everyone talks about. Turns out they can. Most of what gets locked behind a SaaS paywall runs fine on a used 3090 if you're willing to do the setup work.
What I Cover
- Local AI — Language models, image generation, video, voice synthesis — all on consumer GPUs.
- Homelab — Self-hosted services, server builds, and infrastructure.
- Automation — Scripts, workflows, and AI agents that do the boring stuff.
- Privacy — De-Googling, self-hosting, and keeping data on hardware I own.
The Hardware
JARVIS — Dual RTX 3090 AI server (48GB VRAM). Runs local language models, video generation, and voice synthesis. Built from used parts for under $2K. The full build story is here.
ATLAS — RTX 4080 workstation (16GB VRAM). Handles image generation, daily work, and occasional gaming. Every image on this site was generated on this machine.
Start Here
New to the site? Start with how I built a dual-3090 AI server for under $2K or how a single RTX 4080 generates images in 7 seconds. Everything else branches from there.
The Site
Built with Astro and Tailwind CSS, deployed on Cloudflare Pages. Markdown in, static HTML out.
Find Me
All the code from these posts is on GitHub. For work stuff, I'm on LinkedIn. The professional portfolio (the boring version of this site) is at zoltanfabry.com.
Get New Posts
One email per post. No spam, no weekly roundups. Just a link when something new goes up.
Stay in the loop
New builds and breakdowns, straight to your inbox. No spam.