How it started: a dusty GPU and 90 days to fill

It started in September 2024. Surgery took me out of work for about 90 days, and I’m a workaholic by nature, so the empty time got to me fast. I filled it with AI.

My gaming PC was strong enough to run models, but I didn’t want my experiments living on it. So I went digging through old hardware and found my GTX 1080 Founders Edition collecting dust in the closet. That became the heart of my first real AI server.

I used Gemini as my guide from day one. The server sat in a back storage room, so I needed to control it remotely. I upgraded both my main rig and the server to Windows 11 Pro so I could connect to it with Remote Desktop, and then I got to work.

Within a few days I had a full plan. Ollama, a free tool for running AI models on your own machine, was running Llama 3 8B, and I had a local chatbot of my own.

There was a catch, though. Talking to Llama wasn’t like talking to Gemini. My little local model was mine, but it wasn’t brilliant. So for a while the irony ran the show: I spent most of my time building local AI infrastructure with a cloud AI showing me the way.

That was the first janky step. It wouldn’t be the last.