Arc Reactor Control and AI Knowledge Library

Galvanix extracts neodymium, one of the rare earth metals behind the magnets in motors, wind turbines, and electronics. How much metal comes out of a batch depends on how the arc is fired.

The challenge

Galvanix's arc reactor ran on analog controls, set by hand. That made it hard to run the same firing schedule twice, to change one thing at a time, or to keep a clear record of what actually worked.

The team's know-how had the same problem. What they had learned about the process lived in notes, files, and people's heads, not somewhere anyone could search.

What we built

We digitized the reactor's controls so the arc can be driven over a serial connection. Instead of turning dials, the team runs firing schedules: programmed sequences they can repeat exactly, adjust one step at a time, and compare from run to run, all aimed at getting more neodymium out of every batch.

We also set up an AI agent for the team, built on OpenClaw, along with a knowledge library it can read from and add to. Procedures, results, and lessons learned go into one place. Anyone can ask the agent a question, and the agent keeps the library current as the team learns.

The result

Firing schedules are now something Galvanix can program, repeat, and improve, instead of something set by feel. And what the team learns gets captured as it happens, in a library their AI agent can search and update.