Notes from production.
Essays on backends, AI infra, and the small decisions that quietly compound into systems worth working on.
Why I wrote a code generator instead of a framework
Frameworks couple your domain to someone else's opinions. Code generators give you the opinions, then get out of the way. Here's how Define works — and why.
Feature flags are a distributed systems problem
Most feature flag services pretend evaluation is stateless. It isn't. Stale cache, split-brain rollouts, and the rule engine nobody thinks about until it breaks.
Semantic caching in production: what pgvector actually gets you
Everyone talks about embedding similarity for caching LLM responses. Few talk about the cache-poisoning edge cases, the distance thresholds that actually work, and why latency isn't the real win.
I regularly share inspired thoughts on backend architecture, open-source discoveries, and the realities of shipping to production. Come say hi.