AI Infrastructure & Reliability Engineer
Infrastructure engineer with 7+ years in SRE and MLOps, now building safety and reliability tooling for autonomous AI systems. CKS/CKA certified, preceded by a decade in FDA-cleared medical hardware where reliability meant patient safety, 10 peer-reviewed papers and 2 US patents. Focused on providing visibility and reliability in non-deterministic systems.
A 3-part guide to serving LLMs on a GPU-accelerated Kubernetes cluster in the cloud using KEDA for auto-scaling and Prometheus/Grafana for visibility. Describes how to: (1) correctly choose a GPU from first principles (weight memory vs. KV cache headroom, derived to concurrent-request capacity), (2) step-by-step guide on how to deploy an inference model on GPU chips in the cloud, and (3) handle unexpected flapping behavior associated with event-driven autoscaling with vLLM. An article series backed by a reproducible deployment of Qwen2.5-7B-Instruct-AWQ on vLLM across A10 nodes.
How to implement a non-deterministic reasoning engine in a system that requires guarantees? K8gents answers through guardrails-as-code: an autonomous Kubernetes root cause analysis (RCA) agent that routes cluster failures through Gemini-powered analysis, but constrains remediation with code-enforced policies—human approval gates paired with OPA Gatekeeper rules create hard boundaries around AI decisions. Diagnostics are exposed via an MCP server (io.github.JDoornink/k8gents on the official registry); see the README for a discussion of failure modes and confidence calibration tradeoffs.
A self-healing multimodal search engine demonstrating the agent-in-the-reliability-loop pattern applied to ML systems. CLIP embeddings + Qdrant power semantic video search ("find a red truck at night"); Prometheus and Evidently AI surface drift; an LLM agent reasons about drift signals and triggers automated retraining via GitHub Actions. Companion to K8gentS — same thesis (LLM as decision-maker inside a reliability contract), different domain.
Static analysis for the agent supply chain. A published Python CLI tool that validates MCP servers and scans AI agent implementations for security vulnerabilities — configurable security levels, CI/CD integration with threshold-based failure conditions, and SARIF output for integration with existing security tooling.