https://ravisinghal.io/ Writing by Ravi Singhal on agentic AI systems, LLM inference platforms, agent evaluation and AI governance. en https://ravisinghal.io/blog/llm-inference-autoscaling-kubernetes https://ravisinghal.io/blog/llm-inference-autoscaling-kubernetes I pointed an autoscaler at GPU utilisation and it scaled down while latency climbed. That's not a tuning problem, it's the wrong number. Notes from putting vLLM on Kubernetes with a 4 GB card and nowhere to hide. Fri, 18 Sep 2026 00:00:00 GMT ravi@ravisinghal.io (Ravi Singhal) https://ravisinghal.io/blog/asymmetric-automation-agentic-adjudication https://ravisinghal.io/blog/asymmetric-automation-agentic-adjudication Approve and reject look like two outcomes of one decision. They aren't — one is reversible and the other costs someone a year. Everything in this build follows from that, including two bugs I walked straight into. Sat, 12 Sep 2026 00:00:00 GMT ravi@ravisinghal.io (Ravi Singhal) https://ravisinghal.io/blog/kubernetes-servicenow-ai-agent https://ravisinghal.io/blog/kubernetes-servicenow-ai-agent A Kubernetes agent that runs an in-cluster Ollama LLM to autonomously detect failures, correlate root causes, and file structured ServiceNow incidents — no API keys, no hard-coded thresholds, one command to run locally. Thu, 19 Feb 2026 00:00:00 GMT ravi@ravisinghal.io (Ravi Singhal) https://ravisinghal.io/blog/journey-of-a-staff-engineer https://ravisinghal.io/blog/journey-of-a-staff-engineer A look back at two decades of software engineering — from writing my first Java class at HCL to architecting Wed, 18 Feb 2026 00:00:00 GMT ravi@ravisinghal.io (Ravi Singhal)