Ring radar: a senior-judgment one-pager
Four years as TPM in Amazon Ring's R&D org. The headline program: a radar camera taken from sensor research to 800k-2M units shipped -- 10+ teams, 5 countries, spanning ML, hardware, firmware, and software at the same time. This page is not the resume. It's the calls I'd want a hiring manager to ask me about.
One program, end to end. Every box was somebody's whole department.
The scale part
Motion Settings 2.0 went to 40M+ Ring devices, rolled out in controlled waves of 1M a day. On the same deployed hardware, Person Detection accuracy went from 30% to 46% -- measured on devices already on people's walls.
Alongside it: ANVIL, a compliance overhaul that split the monolithic Ring Research AWS account into 25 independently certified services against Amazon's highest data and privacy standards. Zero downtime during the split. Five certifications landed 30 weeks ahead of schedule. Compliance work never makes the highlight reel, which is exactly why I'm putting it here -- it's the least forgiving program type I know.
In total the role delivered a portfolio of 7 products fusing ML, hardware, firmware, and software domains.
What I killed, and why
Two decisions from those four years say more about how I operate than any shipped number.
The rollback. A Motion Settings deploy that was designed to go out in controlled 1M/day waves went global by accident. The feature itself was mostly fine; the deploy was not, and "mostly fine" across 40M+ devices is not a risk you carry while you debug. I managed the full platform rollback, ran the correction-of-errors process, and re-shipped the feature correctly (in waves this time, as designed). Rolling back something that largely works is a harder call than rolling back something on fire -- there's always a voice saying leave it, it's probably OK.
The termination. I made the call to let go a high-performing but toxic engineer. Individually strong output, corrosive effect on everyone around it. After the exit, team cohesion and delivery predictability measurably improved. I'm keeping the details off a public page deliberately (no names, no war story) -- the point is only that I've made the uncomfortable personnel call, and that the team got better when I did.
Both calls destroyed something in the short term: a live rollout, a strong performer's seat. Both were right. Senior judgment, as far as I can tell, is mostly the willingness to pay a visible short-term cost for a boring long-term system that works.