Operator-side measurements of how AI systems perform, scale, fail, and cost—with the method, evidence, and asterisks shown.https://abhijitmazumder.com/en-inhttps://abhijitmazumder.com/posts/how-an-llm-generates-one-token/https://abhijitmazumder.com/posts/how-an-llm-generates-one-token/A distilled understanding of what happens when an LLM generates one token, with an interactive explainer.Fri, 11 Sep 2026 00:00:00 GMTLLMsExplainershttps://abhijitmazumder.com/posts/when-the-benchmark-waits-for-the-server/https://abhijitmazumder.com/posts/when-the-benchmark-waits-for-the-server/Why a closed-loop load test can understate the queueing real arrivals experience, and how that changes my next AI-serving experiment.Wed, 05 Aug 2026 00:00:00 GMTMethodologyAI InfrastructurePerformancehttps://abhijitmazumder.com/posts/who-is-your-benchmark-for/https://abhijitmazumder.com/posts/who-is-your-benchmark-for/“Two defensible answers to 'how fast is this box?', and whose question each answer serves.”Mon, 27 Jul 2026 00:00:00 GMTBenchmarkingPerformanceSystems