I started in web engineering and moved quickly toward the parts of the stack where correctness compounds: data models, service boundaries, queues, caches, and the observability that tells you the truth at 3am. Over 4.5+ years I've taken three SaaS products from prototype to paying, multi-tenant production.
“Scalable systems” in my work means capacity that is a config change rather than a rewrite: stateless services, idempotent workers, explicit backpressure, and load-tested assumptions written down as SLOs.
“AI-first” means the LLM is a component with a contract, not a magic layer — retrieval you can inspect, evals in CI, structured outputs, cost and latency budgets per request, and graceful degradation when the model is wrong.
I work best with product-minded teams across time zones: written-first, small PRs, decisions in ADRs. Outside work I mentor junior engineers and rebuild my homelab more often than I should.