I got my first real look at AI from the receiving end of it. I worked a call-center job at TELUS, in customer service, watching AI get pushed onto customers who didn't want it: automated answers, redirects, menus that led nowhere. The technology was genuinely capable. It was just aimed at the wrong target, used to cut headcount, deflect calls, and treat support as a cost instead of a job worth doing well.
That experience shaped how I think about this stuff. The promise everyone sells is "replace the human." What I saw was that doing so makes the product worse and the customer angrier, because you lose the judgment and context that made the work valuable in the first place. The version of AI I believe in is the opposite: a force-multiplier that makes one person able to do more and do it better, not a cheap way to remove them.
So that's what I build. Local voice and wake-word models, ONNX deployment, tooling around AI agents. Concrete things that ship and earn their keep, rather than another demo riding on the promise of replacing people.
wakeword-forge: local wake-word training -> ONNX exportokay-hermes-repcnn-onnx: compact "Okay Hermes" wake-word modelokay-hermes-voice: always-on local voice daemon for Hermes Agenttier-1-2-3-skill-system: when to use instructions, scripts, or MLrecursive-agent-improvement: repeated agent failure -> specialist toolhermes-linux-ricing: desktop changes with audit/apply/rollback