Have you ever wondered what it would take to build a portable, off-grid AI server that can run on solar and hand-crank power? Well, I did just that, and I’m excited to share my project with you. I call it the SERVE-AI-VAL Box.
My goal was to create a handheld, ruggedized device that could run at least one large language model (LLM) or multiple LLMs at an acceptable generation speed. It had to be capable of running without any connectivity, after initial provisioning, of course. And, it needed to accept multiple power sources, like solar, hand-crank, and AC/DC power.
After some research, I decided to use an Android mini tablet, the XREAL Beam Pro, as the brain of my device. It’s the perfect form factor and can readily run LLMs. For power, I found a Lithium battery power bank with a solar panel, hand crank, and multiple power output types. It even has a bonus flashlight and is ruggedized and waterproof.
I’ve created a GitHub repository with the full parts list, pictures, and instructions for assembly and software setup. It’s not a super complex project, but it was a fun challenge, and I hope it inspires others to explore the possibilities of portable, off-grid AI.
The potential applications of this device are vast. Imagine having access to medical or survival knowledge in emergency off-grid situations where you have no power and no network connectivity. A local LLM could provide critical information in these situations.
If you’re interested in building your own SERVE-AI-VAL Box, I encourage you to check out my GitHub repository and give it a try.