most recently at meta (summer 2026). i fine-tuned an 8b llm to replace an 80b classifier for whatsapp paid messaging — 20% better accuracy at a tenth the size — and shipped typed template parameters that cut pricing-tier abuse 65%.

before that, state farm (summer 2025), where i built api security scanning tools adopted across most of the company's internal apis with terraform and ci/cd. and king's hawaiian (2024-2025), where i built ml models on 11m+ data points to optimize bread production.

i also do competitive programming (icpc north america championship finalist) and like building stuff that helps people, like schedulehero. interested in backend systems, infrastructure, and solving hard problems. i play badminton too.