Cloud platform architecture
Cloud-native, multi-tenant platforms meant to be built on, not around: shared libraries, a common component system and a reference app, so a product team starts from something that already works.
Sydney, Australia
Anurag Pathak
Enterprise architecture, infrastructure, automation & AI.
By day I design and lead enterprise software platforms. Outside work I build and run my own infrastructure — a Proxmox cluster and a 12U rack at home, a server that hosts everything I publish, and an automation pipeline that reads the news and messages me when it finds something worth covering. Most of what I know came from running the thing, not reading about it.
01 — About
What I do
I lead product development at a Sydney software company, after twelve years as its enterprise solution architect.
Outside it I build things that run continuously and look after themselves: self-hosted infrastructure, automation that removes a manual step, and applied LLM work where the cost is measured rather than guessed.
The way I work matters more to me than the tools. Few moving parts. Write down why, not just what. Reason about the security posture instead of assuming it. Understand the thing before installing it.
- Rack
- 12U, 3-node Proxmox cluster
- Pipeline
- 12 feeds, 3 runs a day
- Experience
- 22 years, since 2004
02 — Work
At work
Twenty-two years in software, described in general terms. The specifics belong to the people I do it for.
Kubernetes delivery
One chart template behind every service and environments promoted through declarative config, so adding a service is a values file rather than a pile of YAML.
Supply chain & identity
Base images patched against scanner findings before anything ships on them, and one identity layer across every product instead of one per app.
Observability
The LGTM stack with OpenTelemetry: metrics, logs and traces tied together end to end, so an incident starts with an answer rather than a search.
CI and migrations
Self-hosted build runners that scale with demand, and moving source control and pipelines between platforms without losing a commit, branch or tag.
Engineering with AI
AI-assisted development as a team practice: each codebase carries its own written rules, so the assistant works to the same standards as the people.
03 — Homelab
At home
Things that run continuously, not demos. All self-hosted and operated by me.
AI news-to-video pipeline
self-hosted · designed, built, operated
Polls twelve RSS feeds across four topics three times a day, removes duplicates, scores each story against a per-topic brief with an LLM, writes short-form scripts, and sends them for approval before anything moves. Every model call is costed into a daily rollup — a pipeline running three times a day without a cost view is a surprise waiting to happen.
The homelab
12U rack · started as a spare laptop
Home Assistant on an old laptop grew into a three-node Proxmox cluster, shared TrueNAS storage and a network segmented by design. It is where everything else gets built and broken first.
This site & the blog beside it
self-hosted · designed, built, operated
Two sites on one machine I run and maintain myself: this page, and the blog it pulls its posts from. Built so the next site costs minutes rather than an afternoon — certificates, routing and isolation are solved once, not per site.
05 — Stack
What I work with
Things I actually run, not everything I've touched.
Infrastructure
Proxmox, Docker, Traefik, TrueNAS, UniFi, Ubuntu
Automation
n8n, Cloudflare Tunnel, webhooks, Telegram bots
Data
PostgreSQL, MySQL, schema design
AI
Claude API, prompt and cost engineering, RSS-to-LLM pipelines
Platform
Cloudflare, Let's Encrypt, AWS SES, Ghost
At work
Cloud-native Kubernetes, Helm, Spring Boot, Angular, Keycloak, LGTM + OpenTelemetry
Languages
to confirm
06 — Contact
Let's talk
Happy to talk about infrastructure, automation, or anything on this page.
EmailGitHubLinkedIn