Motivation
Why build a home lab when cloud services exist for everything?
The Skills Were There
After a year at SAP since Jun 2025, I’d picked up a solid foundation in infra, deployments, and CI/CD pipelines. At some point it clicked — I actually know enough to do this myself. Not just follow tutorials, but design something real and own it end to end.
2026 also happened to be the year AI-native development exploded. AI-assisted with Gemini made it genuinely feasible to move fast on ambitious side projects without getting bogged down. Vibe-coding is very different from AI-assisted Coding, because it combines technical expertise of humans with the speed of development
The Problem With “Just Use The Cloud”
Using cloud services for everything means:
- No ownership — data lives on someone else’s servers
- Recurring costs — subscriptions add up fast
- No learning — someone else manages the hard parts
I wanted to actually understand how things work at the infra level. DNS, tunnels, storage replication, sync protocols — the stuff that’s usually abstracted away.
What I Actually Wanted
This idea initially sprung up when I was preparing for my CKAD examination, where I was documenting my notes in Notion, and exploring Obsidian
- A personal knowledge base (Obsidian) that syncs across all devices without relying on Obsidian Sync (paid)
- A public blog (garden.iujinwee.cc) that publishes automatically from my vault
- A portfolio site that I host myself, not on some managed platform
- Custom email on my own domain
- All of it accessible remotely, securely, without opening ports on my home network
The NAS with CouchDB handles vault sync. Cloudflare tunnels handle secure external access. TailScale handles the VPN mesh for when I need direct access. Each piece solves a specific problem.
How I’m building it
Tentatively, the home lab runs on the NAS alone — no dedicated compute node yet. The NAS handles storage and runs CouchDB for Obsidian LiveSync.
Hermes Agent acts as the brain, its self-improving agent harness is quite fascinating, but currently exploring how well it performs..!
The actual heavy-lifting is currently dedicated to OpenRouter API, which has free models, so kudos to them! No need to pay for any subscription at the moment & I can test this out :) Will keep using this until they paywall their APIs, but it’s pretty decent for
No doubt, the future plan is to get the M5 Mac Mini, where I can self-host my own local Ollama server which fully removes the dependency on cloud-based APIs for LLM inferencing. But at the moment, I don’t see the dire need and also it’s not even released yet!
So let’s observe for now…
Let’s see how this pans out in the future 👀🥂