Dillon Browne

I'm a Staff Engineer and Cloud Architect — 10+ years of resilient infrastructure at scale: multi-cloud, Kubernetes, IaC, observability, and security. Lately, production AI platform work: agents, LLM serving, GPU orchestration.

Yes. I take on full-time roles and contract engagements, remote or on-site, and typically respond within 24 hours. Use the contact form to start a conversation about your project or role.

Production LLM deployments and the plumbing tool-using agents need: typed tool schemas, capability scoping, context pipelines, GPU orchestration. The hard part isn't what a model knows — it's what an agent is allowed to do. I've shipped MCP servers with OAuth 2.1, least-privilege scopes, short-lived credentials, and audit trails that say which AI system reached which data.

Underneath it. I came in from the offensive side — red team, network exploitation, app and API security — so the threat model comes before the control: what an agent may do, what happens when a tool hands the model hostile text, what an anonymous caller can cost you. The agent on this site is the worked example: every control fails closed, and /work lists each one beside its implementation and its test.

Right-sizing, spot/committed-use strategy, autoscaling tuned to real demand, storage-tier and egress review, and continuous FinOps measurement. Past engagements have delivered roughly 40% average infrastructure cost reduction without sacrificing reliability.

Edge-first: Astro static site with React islands, deployed on Cloudflare Workers with serverless functions, Workers AI for the chat assistant, and KV for rate limiting and caching. Read the full write-up in the "How I Built This" blog post.

From advisory and architecture review through to hands-on implementation and operation — cloud migrations, Kubernetes platforms, CI/CD and GitOps, observability, and AI/infrastructure integration, scoped to your needs and outcomes.