First, I've been touching computers since before DOS, but since I'm a hardware engineer, I'm very weak on the software side.
During college, I saw C++ and chose to focus on hardware instead, and I've been living with that decision ever since.
But now it's the age of AI, and at the company too, there's always this atmosphere of "how much are you utilizing AI", so I couldn't help but use AI as a chatbot. This time, I decided to use a spare computer at home to set up a personal AI assistant, and this is a chronicle of my struggles in the process.
First Struggle
Among the spare PCs at home, the one that caught my eye was Project Voltera, a Dev Box that Microsoft created back in the day to build the ARM ecosystem and sell. It's a model with Qualcomm's older chipset and DDR4 32GB memory. But this machine is absolutely blocked from installing anything like Linux. Only Windows is possible. So I boldly installed WSL and Ubuntu.
I opened up SSH, installed Tailscale, installed Hermes, registered APIs and all that... I set everything up diligently with the help of Gemini. Connected it to Telegram too... Now I should try using it, but ugh... Windows updates are getting in the way. It keeps rebooting out of nowhere, and it's so annoying.
So today, I started thinking about whether there's a way to run this in a Linux environment.
What else could I run at home? Looking around... Then I noticed a machine on my desk that's been running lightly without doing much heavy work. Bee ME Mini. It's configured with an N150 and can accommodate up to 6 SSDs. I had TrueNAS installed on this machine. When I asked Gemini, they—who previously opposed running an AI machine on this device—agreed it was a good idea. When I asked why, they said running something like Ollama would be difficult, but running something like Hermes is no problem. They just convinced me to outsource the inference. Anyway, the previous PC had only a decent amount of memory but weak GPU, and the fan would spin like crazy if I tried to run anything on CPU power, so I had already decided to outsource inference. So I decided to go that route.
Fortunately, Gemini recommended not installing from scratch but migrating instead, so I packed up all the files I had previously set up and moved them over to TrueNAS.
So what am I doing with it now... Well, experts might wonder why I'm running something like that, but I'm planning to have it send alerts when emails come from my child's school, and since meal planning is annoying, I'll tell it what I've been eating, and once it accumulates enough data, I'll have it extract next week's recommended menu suggestions every Friday evening so I can shop on Saturday. When I watch YouTube tutorials and hear how companies use these things and talk about such impressive applications, I'm actually building a completely lifestyle-oriented AI machine, but at over 50 years old, it's so much fun—like discovering a fun toy.
Even a complete beginner like me is having so much fun thanks to AI, installing things through trial and error, and thinking about applications together.
Please be kind if I ask naive questions later.
