Tokyo Calling

Sometimes the universe gives you a sign. For me, it was a cat cafe in Tokyo, a bullet train ride that felt like floating, and a city that operates with the kind of precision I've spent my entire career trying to build into software systems.

And just like that, we're moving to Japan.

The Decision

After 8 years and 8 months at CrowdStrike, I made the call to leave. Not because of any dramatic moment, not because of any particular project or disagreement. Just a realization that it was time to pursue something bigger.

Sometimes you build enough infrastructure, automate enough systems, and stabilize enough servers to know when the next chapter is calling. For us, that call was Japan.

Quantum Mesh

I'm joining a think tank based in Tokyo as a Distinguished Engineer. The work will span technical leadership across a variety of initiatives. Think tanks move at a different pace than product companies — there's room for deep thinking, for building systems that matter without the pressure of quarterly roadmaps.

We're still building software. Still building systems with AI. Still automating everything that can be automated. Just with a different backdrop. A Tokyo backdrop.

The Data Stack Work

Before the big move, there was one last piece of infrastructure to build out. The requirement was simple: connect arbitrary data stores to a compute plane for arbitrary joins.

Enter Trino.

If you've never configured Trino, it's refreshingly straightforward. The architecture is clean:

Data Sources → Trino (Query Engine) → Superset (Visualization)
     ↓
   Hive (Storage Layer)

We built Ansible playbooks to deploy the entire stack. Because of course we did. At this point, if it's not codified in Ansible, did it really happen?

The aha moment: Trino's catalog-based configuration model means you can add a new data source by dropping a properties file into a directory. That's it. No complex orchestration, no Kubernetes operators. Just a config file that says "here's where the data lives."

It's the kind of simplicity that reminds you why we got into this work in the first place.

The Toaster is Stable (Finally)

Remember the toaster? Our 4-GPU Blackwell beast that was crashing every single day? Yeah, that saga finally has a happy ending.

Turns out the original ASRock power supplies were the culprit. We replaced them with Corsair Prime 1600W units, and the system has been rock solid for weeks now. All the PCIe slot temps are stable, the GPU power limits are holding at 250W, and the fan control daemon is keeping everything cool.

The full stability stack is still in place:

  • Kernel params: pcie_aspm=off pcie_port_pm=off
  • GPU power limit: 250W
  • Aggressive fan curve via gpu-fans service
  • Continuous monitoring via pcie-monitor

The toaster is finally tamed. And it's coming with us to Tokyo.

What's Next

I'm departing at the end of the month. This is intended to be a permanent move, though we'll see how permanent it actually is. There's still the small matter of learning the language to figure out.

The work at Quantum Mesh is a culmination of everything I've been building toward for the last decade. Eight years of infrastructure, automation, and AI-accelerated development — all leading to this moment.

I hope I can bring those skills to the team and build something amazing.

The Journey Continues

This blog has documented a lot of projects over the years. Server builds, AI experiments, code that makes computers do impressive things. But this? This is the biggest project yet.

Building a life in a new country, with a new team, in a completely different context.

Wish us luck. We're going to need it.

See you in Tokyo.

This blog post was written with the help of Qwen 3.5.