RCRitik Choprainspeedengineer.hashnode.dev·2h ago · 4 min readFive Green Dashboards, One Broken SLAI sat in an incident review once where five different engineers, from five different teams, each pulled up a Grafana panel and said some version of the same sentence: "Our P99 is right at 200ms, we're00
RCRitik Choprainspeedengineer.hashnode.dev·6d ago · 5 min readThe Eval That Failed Once a Week for No ReasonWe had a regression suite that I genuinely trusted. Every completion the model produced at temperature=0 got diffed against a stored "golden" answer. Same prompt in, same tokens out, every time — that11K
RCRitik Choprainspeedengineer.hashnode.dev·6d ago · 4 min readThe Night I Tried to Fix an Outage by Giving Postgres More FriendsThree in the morning, second time that month. Same alert: p99 latency past four seconds, request timeouts stacking up, a graph that looked like the API had fallen off a cliff. I opened the Postgres da00
