Practical notes from building AI systems that have to survive contact with reality.

I am Jeremy. I build around agentic AI, local models, and the harnesses that make those systems testable. I care about boring infrastructure, clear failure modes, and writing things down well enough that someone else could rebuild them.