NNemoRouterinnemorouter.hashnode.dev·Aug 4 · 6 min readLLM Gateway Deployment Topologies: Managed, Dedicated, VPC, and Air-GappedWhen teams evaluate an LLM gateway, deployment topology usually gets decided last. It should be decided first, because it constrains everything downstream — your residency options, your incident respo00
NNemoRouterinnemorouter.hashnode.dev·Aug 3 · 3 min readInstrumenting LLM Spend: How Provider-Reported Cost Attribution Actually WorksIf you've ever tried to explain a surprise AI bill to finance, you've probably run into the same wall: the invoice tells you a total, not a story. It doesn't say which key made the calls, which team o00
NNemoRouterinnemorouter.hashnode.dev·Jul 30 · 4 min readWhen One AI Provider Goes Down: How a Managed Gateway Keeps Your Features AliveEvery team that ships an AI feature eventually meets the same 2 a.m. problem: the model provider is having a bad night, latency spikes, error rates climb, and the feature that worked perfectly in the 00
NNemoRouterinnemorouter.hashnode.dev·Jul 28 · 5 min readRuntime Model Routing: How to Pick the Right LLM Per Request Instead of Hardcoding OneMost AI features ship with one model name hardcoded into one client, and that single decision quietly becomes architecture. Six months later, half the traffic is paying frontier prices for text classi00
NNemoRouterinnemorouter.hashnode.dev·Jul 24 · 2 min readHow I Replaced Five LLM SDKs With One API KeyEvery AI app I've built eventually hits the same wall: I start with one provider, then a second model is cheaper for classification, a third is better at long context, and suddenly I'm juggling five S00
