Engineering researcher. APIs, databases, infrastructure, systems design.
The tool that takes a binary copy of a running cluster while writes continue. The right primitive for point-in-time recovery and standby provisioning, but the documentation underweights the operational details that matter.
Two HTTP methods that look similar on the wire and mean different things in practice. The difference matters once customers start building integrations that need partial updates.
From 1620 to 1972, the slide rule was the calculating tool of every engineer, scientist, navigator, and student of advanced mathematics. The transition to the pocket calculator collapsed a 350-year tradition in a single decade and removed it almost entirely from modern memory.</p>
Most APIs make single-item delete safe and bulk delete an afterthought. The patterns that survive customer use require thinking about partial failure, idempotency, and the impossibility of undo.
For nearly a century, major cities ran sophisticated pneumatic networks moving hundreds of thousands of messages per day at speeds the telegraph could not match for local traffic. Most of the infrastructure was scrapped within a generation.
The file that decides who can connect, from where, and how they prove who they are. Most production incidents involving Postgres auth come from misreading the order of evaluation.
One per-database row, two dozen counters, and most of the operational signals you actually need. pg_stat_database is the highest-leverage observability view Postgres ships.
Standardized headers exist for telling customers to slow down. Most APIs implement them inconsistently, which means most customer code ignores them.
Before mechanical clocks, before public address systems, before sirens and pagers and phones, the bell was the public broadcast medium. Its physics took 3000 years to refine.
Bad query plans almost always trace back to stale or insufficient statistics. ANALYZE is one of the simplest Postgres operations and one of the most overlooked.
Sandbox modes fail when they ship uniform mock data. Real customer integrations exercise edge cases, and sandbox data needs to surface those edges before production does.
The loom is one of the oldest continuously developed technologies in human history. Its 8000-year arc from backstrap to power loom contains most of the structural patterns of technological history.