I build the quiet machinery behind money that moves: the payments, and the prices you watch tick on a screen. Then I go home and keep a small cloud humming in the corner of the room.
I grew up in Sukabumi, live in South Jakarta, and I'm drawn to the unglamorous details, the ones nobody notices until they break. Taking those slowly and with care is where I find the good work, and most of the fun.
Backend and platform work in fintech, mostly in Java and Go.
- Live market data. Every price and trade from the exchange feed, parsed and pushed to every open app in real time, over Kafka, Hazelcast and WebSockets. Lately: moving all of it onto a self-service Kubernetes platform, with no regressions so far.
- Products from zero. A brokerage product built on a partner's API, from accounts to the full order lifecycle to corporate actions, live in under three months.
- Money that has to add up. Payment rails into local banks (virtual accounts, cards, direct debit) and the reconciliation systems that keep every unit and every rupiah exact.
Building things with friends, from small products to the teams behind them.
- Agents that do the paperwork. Fine-tuned models that write complete documents, customer-facing agents, and a bot in our chat that hunts bugs alongside us. Self-hosted or on a provider, whichever fits the job.
- Weather at planet scale. Climate data from Copernicus, WRF simulations on our own Kubernetes and Proxmox, and results delivered through a web app and editor plugins.
- A home lab. Two Proxmox hosts, one k3s cluster, three tenants. Declared in git, synced by ArgoCD, reached only through the edge; nothing listens on the home IP. My notes, code and agents all live there, and isnan.me can run a live speed test all the way into it.
The agent drafts. I own the diff.
I hand AI the context on purpose, let it write the feature, then read it the way I'd read a stranger's pull request: logic first, then queries and indexes, then memory, storage and CPU. It goes through a dev container and CI before it ships. I build my own MCP servers and fork open-source tools to wire AI into them, and none of it gets trusted blindly.
I also write almost everything down, mostly for my agents: decisions, runbooks and incidents go into a wiki they read before they write a line of code.
Java · Go · Python · Kafka · Kubernetes · k3s · Proxmox · ArgoCD · Terraform · Cloudflare · MCP · Ollama
Away from the keyboard, I sing. I learned to read a score in Paragita, the Universitas Indonesia student choir, and these days I arrange as much as I perform. I listen across genres and follow the music more than the artist: RAYE lately, Tulus closer to home, and Sia, always.
| Isnan's Jam What's on repeat. Press play on Spotify. |
- Since 2021 I've taught programming at my old high school's IT club back home. Tenth graders start with Python and algorithms; by twelfth grade they're shipping Django and Flutter apps. Still my favourite way to learn something twice.
- I read other people's C# for fun, mostly Cities: Skylines mods, to see how they bend an engine that was never meant to be bent.
- The next language on my list isn't a programming language. つぎは、日本語。
- The agent drafts. I own the diff.
- A silent feed looks healthy
- Reconciliation is a product, not a script
- A home lab that runs my second brain
- What teaching tenth graders taught me about code review
More at docs.isnan.me, the public side of the wiki my agents read.
Open to hard problems in backend, infrastructure or platforms; to building something with agents in it; and to teaching anyone who wants to learn.






