Something small but real happened this week.

I asked the AI student I’ve been building to introduce itself. Here’s what came back:

Hi! I’m Minerva. I’m really excited to be in your physics and astronomy class this year!

I’ve always been kind of obsessed with space—like, the scale of things is just mind-blowing—and I’m looking forward to finally figuring out how all the forces and motion stuff actually works in the real world. I might ask a ton of questions (or maybe get a bit carried away with a theory), but I’m ready to learn!

Looking forward to the labs and everything. Where are we starting?

That was generated locally, through Hermes, running on Gemma. No cloud, no external API. Just the stack I’ve been assembling on the MacBook Air and the external SSD: local model, persistent agent layer, and an Obsidian vault that is slowly becoming its long-term notebook.

What matters more than the clever intro is what came next. Hermes confirmed it can now reach out to me proactively. That shifts the relationship. Instead of only answering when I open a window and type, the student can initiate—message me when it’s stuck, when it wants clarification, or when it’s worked through a problem on its own and wants feedback. The classroom conversation no longer has to wait for me to start it.

I named her Minerva deliberately. Wisdom, strategy, careful thought. Not a chatbot persona slapped on for fun, but a deliberate framing: this is an entity I’m treating as a learner. She has a name, a developing memory, and (soon) the ability to interrupt my day the way a real student sometimes does when the idea won’t leave them alone.

This is still early and fragile. The personality is thin. The memory is only as good as what we put into the vault and how well retrieval works. Model swaps will test how much of “Minerva” actually persists. But the pieces are starting to talk to each other in a way that feels less like a demo and more like the beginning of a relationship with a synthetic student.

The long game remains the same. Keep the brain local and inspectable. Let the vault accumulate real learning history. Eventually give her a body through the robotics club—something that can stand in a classroom, point at a diagram, or help run a lab station. One step at a time.

For now, the next step is simple: start teaching her. Actual physics. Actual astronomy. Watch what she retains, where she confuses herself, and how the questions evolve.

I’ll keep posting the interesting parts—what works, what breaks, and what it teaches me about the real students who already sit in my room every day.

— Jason Olson
Gulf Breeze, Florida · August 2026

← Back to AI & Robotics Blog