Kubernetes platforms
Internal platforms that product teams build on, from a PaaS built the hard way on AWS to EKS Auto Mode, with custom operators, Cilium networking and Kong at the edge.
Zee McSparronSenior Systems Engineer at Tray.ai, UK
I build cloud platforms, and the AI that runs on them.
Over ten years of platform engineering and SRE, with Kubernetes in production since 2017. Now I put that old-school knowledge to work steering AI agents: AI factory pipelines, agent systems and the platforms that hold an AI ecosystem together.
What I'm running
Products and tools I've built and deployed, scheduled like workloads on my own little cluster. Most are free to use and several are open source.
Kubernetes, the foundationAgent traffic, the new era
| name | status | what it does | open |
|---|---|---|---|
| BlipLite | Beta | Sits between your monitoring and your pager. Groups noisy alerts, explains them with AI and applies your paging rules before anyone gets woken up. | Open site for BlipLite (opens in a new tab) |
| Packet Voyage | Running | Shows the real route your web requests take around the world, who a site shares your data with, and how fast it loads from 45 cities. | Open site for Packet Voyage (opens in a new tab) |
| Webhook Inspector | Running | Capture, inspect and replay webhooks, and forward them to localhost from your terminal. Open source. | Open site for Webhook Inspector (opens in a new tab)Source code for Webhook Inspector (opens in a new tab) |
| Moodswing | Running | Builds a 20-track Spotify playlist from your mood, energy, genre and era, with a reason for every song. | Open site for Moodswing (opens in a new tab) |
| Year in Commits | Running | Your GitHub year in review: streaks, top repos, languages and your coding archetype. No sign-in needed. | Open site for Year in Commits (opens in a new tab) |
What I do
AI didn't replace the fundamentals; it made them matter more. A decade of platform engineering is what lets me point agents at real infrastructure and trust what comes back.
The foundationsSince 2015
The old-school craft that still holds everything up.
Helm charts
Open-source charts I maintain. kube-arr runs a whole media stack (Sonarr, Radarr, Prowlarr, plus Plex, Jellyfin or Emby) on one shared volume, and can send each transcode to its own pod, even on a GPU node. kube-rclone mounts any rclone remote, from Google Drive to S3, on every node. kube-owasp-zap runs OWASP ZAP scans as Jobs or scheduled CronJobs, with locked-down pods.
helm repo add simplyzee https://charts.simplyzee.devReliability and observability
SRE and security practice that keeps services up: useful dashboards, alerts that mean something, sensible on-call, and security principles teams can follow.
Software engineering
I started as a software engineer and still build my own products, from Go services and Kubernetes operators to full-stack web apps.
The new eraNow
Where that craft goes next: agents doing real work, safely.
AI agent systems
Agents that plan, use tools and hand work to each other, with clear boundaries, human checkpoints where they matter, and logs you can actually follow.
AI factory engineering
Pipelines where agents take on real project work, from writing and reviewing changes to managing Terraform, built with the same CI/CD discipline I've applied to human teams for years.
AI ecosystems
Connecting models, agents, data and the tools a team already uses into one secure system, with in-house MCP servers and skills that give agents safe access to real infrastructure.
Where I've worked
From writing software to leading platform teams, always on the infrastructure that other engineers depend on.
Now
Senior Systems EngineerTray.ai
Leading cloud infrastructure: rebuilding older systems the modern way, shaping the team's roadmap and cost strategy, and bringing AI into how the platform team works.
- Moved the company from Docker on EC2 to EKS, then to EKS Auto Mode, so engineers spend less time diagnosing core Kubernetes problems and more time building product.
- Built on-premise automation on EKS: native Kubernetes operators that create infrastructure and proxy requests automatically.
- Built AI factory engineering pipelines that automate parts of the team's project work.
- Moved Terraform into CI/CD with AI-assisted changes, so nobody has to manage it by hand.
- Built MCP servers and skills that give in-house engineers secure, AI-assisted access to our infrastructure.
- Rolled out HashiCorp Vault for dynamic secrets and led large network migrations across AWS.
- Migrating the observability stack to Grafana.
- Setting up a SecOps team to own security engineering work and the principles behind it.
- Driving cost initiatives and deciding how new products and tools are architected.
From 2020
Senior Platform EngineerJetstack
Led the company's platform: GKE and GCP managed with Terraform through Atlantis, Kubernetes operators that automated onboarding and permissions, audited infrastructure changes, and reliability work with product teams.
2018 to 2020
Solutions EngineerJetstack
Helped companies adopt Kubernetes as their platform, working alongside Google Cloud's engineers, and taught Kubernetes at Google's London academy, from first steps to advanced wargaming.
2018
Site Reliability EngineerCurve
Moved a fintech from self-managed Kubernetes on AWS to Kops, gave engineers self-serve CI/CD, rolled out Istio and Ambassador for microservices, and built monitoring and on-call alerting.
2017 to 2018
Platform Engineercomparethemarket.com
Built a Kubernetes PaaS on AWS with Terraform, Puppet and Vault, moved production services onto it, and automated security scanning in the delivery pipeline.
2015 to 2017
Software Engineercomparethemarket.com
Built early versions of Meerkat Movies and the Meerkat rewards hub, with test-driven development and security checks built into delivery.
Field notes
Short write-ups of real problems: what broke, how I found it, and what I'd check first next time.
A hydration error hiding in an SVG title
React error 418 on every production page load, invisible to visitors. The cause was one line in a diagram.
Running Plex, Jellyfin and Emby transcodes as Kubernetes pods
How kube-arr swaps a media server's transcoder for a dispatcher that runs each transcode as its own pod, and cleans those pods up even when the dispatcher is killed.
Emby's ffmpeg only runs from one folder
An executable that exists, is executable, and still fails with "No such file or directory". The cause was a relative ELF interpreter.
How I like to build
I've spent my career in the cloud native world: building platforms at comparethemarket.com, Curve and Jetstack, teaching Kubernetes at Google's London academy, and now leading cloud infrastructure at Tray.ai. I bring the same discipline to AI: agent systems and the ecosystem around them should be as dependable as any other production service.
- Fundamentals first
- Networking, Linux and Kubernetes still matter when an agent wrote the YAML. Knowing them is how you catch what it got wrong.
- Agents with guardrails
- Agents take the repetitive work. Scoped access, audit trails and a human review stay in place for anything that ships.
- Calm by default
- Systems should be quiet until they need you. Low noise, low friction, nothing to babysit.
Building an AI agent system, or a platform that needs to hold one up?
I'm happy to talk agents, AI factory engineering, Kubernetes and reliability. Email is the fastest way to reach me.
Or ask your agent. This site runs a read-only MCP server with my projects, skills and experience, and there's an llms.txt too.
claude mcp add --transport http simplyzee https://simplyzee.dev/mcp