World models built On human expertise.

Organizational World models.

Alexandria Lab builds on what our deployments record: how experts really work at their desktops, across every application, hour after hour.

It is the data long-running AI has been missing, and it stays inside each company.

We build with the people whose work we learn from.

Each person owns the record of their own work and decides what is shared. Training and inference run in secure enclaves, so the data stays with the company it belongs to.

Alexandria is backed by the inventors of Palantir Foundry, the founding team behind Sam Altman’s World.org, and engineering leaders from Salesforce and Uber.

We are competitive programmers, math olympians and physics PhDs from TUM, LMU, ETH and Caltech.

Research here ships.

Every result goes into deployments at companies in regulated, IP-heavy industries. We run experiments on our own GPU clusters and carry the ones that work into production.

Environments & evals How do you measure any model on a company’s real work?
Action models What should an AI do next inside an expert’s software?
World models What will an action lead to, weeks from now?
Credit assignment Which steps earned an outcome that arrives weeks later?
Continual learning How does a model keep learning from each new day without forgetting?
Environments & evals How do you measure any model on a company’s real work?

Open role in the lab