How I work

I would rather ship something real and find out it is wrong than plan something perfect and never find out at all. Most of what I have built started as a weekend deadline, and the habit stuck: get a working version in front of the problem early, then let what breaks tell you where the real work is.

I move between machine learning, backend and hardware without treating them as separate jobs — the same week has involved training an evaluation network, wiring a search pipeline, and soldering a robot that had to carry a kilogram around a track. I also talk to the people I am building for, which engineers are often happy to skip: 20+ onsite interviews, 5 product sales initiatives, and 50+ sales emails while getting Filament off the ground. The fastest way I know to stop building the wrong thing is to ask someone.