It started as an after-hours project during the pandemic. I was using Apache Tika to text-mine an archive of PDFs and scans, looking for patterns. Tika runs on Java, and I didn't want to clutter my laptop with it, so it went into a VirtualBox. The Vagrant box grew into a cloud-init script, then a Packer image, then a Terraform stack. When I wanted to show results to people who weren't developers, ngrok felt wrong, so I put Apache Guacamole in front of it. Guacamole only pipes desktop audio one way, from the remote host to the browser: no microphone back-channel, no conference. So I bolted Mumble alongside. The shared cursor, the thing that now defines the product, fell out of the way x11vnc exposes the X11 screen: I didn't design it, I noticed it.

First I used the stack myself, then friends and family became test pilots. Now it's a product for teams that hit the same seam.

That this grew into a real product at all comes down to an unusual window. I'm a data scientist at Lufthansa zeroG, the group's data and AI unit, and when air travel collapsed during the pandemic I spent more than a year on Germany's Kurzarbeit Nullprogram: still employed, but with no work assigned. What had been an after-hours experiment turned into full-time work on Flexpair, long before anyone was paying for it. My employer knows about the project and is fine with it.

Flexpair is a commercial product. Our business model is managed onboarding, support, and compliance work for teams that want the benefits without owning the Terraform lifecycle themselves — on top of a flat-rate hub fee. You bring your own cloud account; we keep the product running.

Flexpair is built for data scientists, ML researchers, and AI engineers who still, in 2026, spend too much time asking “can you see my screen?”. Flexpair answers: yes, and we're already inside it together.

The team is still small. On the test drive you'll talk to me directly, not a chatbot, not an SDR.