I've spent 6 years at Meta Reality Labs owning what happens between a user hitting a problem and an engineer fixing it. I build the tools for it too.
Senior QA & Product Ops at Meta Reality Labs. Currently building pointd, Interview Prep Kit, and qa-agent.
Points redemption is too complex. Most people sit on points for years — they know they're valuable, but the math and research push them back to cash. pointd skips the complexity and shows you where your points can take you, across every program you have.
Prepping for several interviews at once means rebuilding the same context in a new chat every time. Interview Prep Kit keeps your resume, each job description, and your stories in one place, then generates the prep docs, flashcards, and audio coaching from them. Local-first, open source, bring your own key.
Building fast with AI creates coverage blind spots — tests pass but the right things aren't being tested. qa-agent runs on every PR and flags gaps before they merge, so nothing slips through.
Automated testing catches regressions. It doesn't catch the UX issues users actually notice. A multi-bot pipeline runs on every automated pass, deduplicates candidates against open bugs, and verifies what's left in simulator before it reaches a human.
runs
analysis
filter
verify
review
Designed and consolidated the automated regression suite for a consumer mobile app on Android and iOS, executed by AI test agents across real devices, emulators, and simulators. Roughly 250 test jobs a day, with every result triaged into product bug, test drift, or infrastructure block.
Traced a long-standing "access denied" error that was silently blocking a large share of the iOS automated suite, through the account-provisioning service and identifier mapping down to a single incorrect constant. Restored the blocked coverage, and the same analysis disproved a reported quality regression as sampling variance rather than a real product change.
Extended the provisioning service to support feature-flag enrollment by default, minor and youth account pairs, and propagation-aware waits. Unblocked whole categories that had never been testable: age-restricted flows, parental controls, and onboarding paths.
On-call quality owner across roughly 180 releases: triaging incoming defects, verifying launch-blocking fixes, rerouting mistriaged reports, and signing off releases. 6,000+ distinct defects worked, 4,500+ of them triaged or fix-verified for other engineers.