Lab

Research that ships.

We work on the hard parts of production AI, and test them with clients before anyone else.

What our models seeEveryday moments, read as the variables behind each decision: who will pay, where to open, what is about to fail, what to charge.
  • Causal MLWhy it happens, not just what correlates
  • Multi-agent systemsOne core, many countries and rules
  • Expert reasoningEvery step auditable by a specialist
  • Evals & controlsTested before it touches a customer
Evaluation · oncology research · hereditary cancer

When our AI and the hospital’s commercial tool disagreed, physicians sided with our AI 27 times out of 37.

Sided with our system · 27Sided with commercial tool · 10
525variants
1,152patients
98%expert agreement

Research lines

What we work on.

Signal

Causal ML

Uplift, survival and causal-inference models that say what would change an outcome, not just who correlates with it. Behind our collections, retention and pricing work.

Action

Multi-agent systems

Orchestration, tool use and memory for agents that act in core systems, running across 13 countries on one shared core.

Expertise

Expert reasoning

Agents that gather evidence, apply professional guidelines and write their reasoning down, validated against specialists.

Loop

Evals & controls

Evaluation sets agreed with the business, guardrails, audit logs and EU AI Act transparency tests on live traffic.

Got a problem nobody could crack?

That’s the kind we like. Tell us about it.