Phil Rich
Infrastructure & Agentic Engineering
Networks, Cloud & AI Agents. Co-founder of Quiet Loon.
How I work
I lead infrastructure teams across Europe and the US. Alongside hands-on engineering, I use AI agents to plan, build and review infrastructure changes, and coach colleagues in the approach.
Selected work
-
Built and took to paying customers on Stripe subscriptions. Its core is an asynchronous extraction pipeline that turns inbound messages into structured events, tasks, and reminders, classifies AI failures into transient versus permanent, and retries three times with exponential backoff before surfacing the error to the user.
-
Agentic engineering
Ran a pilot of AI coding agents across infrastructure operations, whose most in-depth output migrated a large next-generation firewall rulebase into Terraform behind a staged GitOps pipeline. I now run an orchestrator agent that briefs parallel worker agents in isolated git worktrees, and track cost, model mix and failure patterns from session telemetry.
-
Local and edge LLM inference
Characterise open-weight models on a 16GB consumer Blackwell GPU against pre-registered decision gates, covering quantisation, speculative decoding drafters and a blinded quality evaluation that decides which model holds the production slot. One investigation traced an apparent build regression to the prompt corpus rather than the software. Extended to the integrated GPU, NPU and CPU of a Meteor Lake laptop, where the OpenVINO NPU path surfaced a reproducible defect that corroborates a known upstream issue.
-
Cloud network architecture
Architected hub-spoke DMZ networks across multiple Azure regions, with per-workload network segmentation, BGP dynamic routing over multi-site VPN and private endpoints for AI services, with every change delivered through a pipeline.
Background
Network engineering, cloud infrastructure and identity, with a recent focus on agentic workflows, local LLM inference and AI governance.