Writing
Notes from the playground.
RSSField notes on system design, algorithms, and infrastructure — the canonical home for posts also syndicated to Hashnode, dev.to, and Medium.
3 posts
sort: newest
- 12 min read
vLLM reinvented the operating system, and nobody told you
PagedAttention is virtual memory. Prefix-aware routing is consistent hashing. Continuous batching is a scheduler fighting head-of-line blocking. The AI infrastructure interview is an operating systems interview in a nicer jacket — here's the map, and the four places it genuinely breaks.
system-designllminferenceoperating-systemscachinginterviewsRead - 11 min read
Why your LLM app gets 429s even when you're under the rate limit
Your dashboard says 40% of capacity. The provider says 429. The gap is that LLM APIs meter a resource whose size you don't know until after you've spent it — and the token bucket you learned in system design prep assumes the opposite.
rate-limitingllmsystem-designdistributed-systemsapi-designRead - 7 min read
Three problems I had to solve to teach algorithms in a browser
What I learned building Subroute — interactive in-browser simulations of system design algorithms. Three engineering problems, three teaching lessons.
system-designengineeringweb-developmentlearningjavascriptRead