Section
Systems
Cloud, reliability, and the craft of running things at scale.
Caching containers
Containers look like ordinary key-value cache entries — until you notice that serving one consumes it. That single difference turns caching into inventory management.
Touch Production with Care
Production access is necessary but risky. A way to score the riskiness of every "touch" — using embeddings and clustering — so owners can act on the riskiest first.
Detecting software regressions
When a fast-moving cloud regresses, which component is the culprit? A detective's approach — spatio-temporal correlation between upgrades and health — built into a tool called Gandalf.