Intelligence is becoming a material you can build with.

It's a strange material. It moves weekly, breaks in places you didn't expect, and asks for a kind of judgment that's still being worked out. Almost no one has a steady feel yet for what it can do, where it ends, or what it costs to use well.

the work

The teams we work with don't need another talk about what AI might do someday. They need one real thing in production, built by people who'll still be there in six months. They have a process worth changing, a product worth extending, or a question worth answering — and a team that could ship it, if they had the right hands in the room for a while.

That's the work.

how we work

We're a two-person studio. We embed with your team for six to twelve weeks and co-build one thing that matters in your domain. Fixed scope, fixed price, fixed timeline. Your team is in the room from day one. Working prototypes weekly, not slides.

method

We work across the stack — retrieval, agents, multimodal interfaces, sensor data, evaluation — but the part we care about most is fit: what should this system actually do, where should it sit, who should it answer to, and how will you know when it's wrong. Most AI work fails at that layer, before any model is chosen.

We build on European infrastructure where the data, the regulation, or the customer asks for it. That usually means Mistral, local deployment, and a stack you can audit end-to-end. Where it doesn't matter, we use whatever ships fastest.

By the end of the engagement, you have two things: the thing itself, and a team that's no longer waiting.

who we are

We met at design school in Potsdam in 2013 and built a brain-computer interface for our thesis. We've been working in adjacent shapes ever since. Synblob is what happens when we stop doing it part-time.