Small, senior teams building load-bearing infrastructure.
The work
We hire engineers and researchers who can sit across the table from the institution that will run the work in production. The bar is craft, judgement, and the willingness to own a result end to end.
Speculative enquiries from infrastructure engineers, AI researchers, and ontology engineers are welcome and read by the team.
Build and tune the kernels and runtime that serve our models in production. The work is low-level: GPU and CPU execution, memory, and scheduling, measured against the latency and throughput budgets the rest of the platform depends on.
Design the compute platform that places, schedules, and scales agent workloads across clusters and bare metal. The mandate is correctness under failure: consensus, isolation, and resource accounting that hold when nodes and networks do not.
Build the control plane that admits, leases, and governs agents as isolated workloads. Think APIs, state machines, and reconciliation loops, with a hard requirement that every enforcement point fails closed.
Build the harnesses and datasets that decide whether a change is safe to ship. The mandate is rigor: reproducible measurement, honest baselines, and evidence that survives scrutiny.
Build the agent runtime and the context that feeds it: tool use, memory, retrieval, and the reasoning loop. You own model behaviour end to end, from prompt to governed action to audited result.
Do original work on how meaning is resolved across an enterprise’s data and how agents reason over it. This is a research role with a publication track and a direct line into what we build.
Sit with the institutions that run our software in production and make it work in their environment. Part engineer, part trusted advisor, you turn a hard deployment into a reference.
Own enterprise relationships from first conversation to signed contract across finance, healthcare, and the public sector. You sell technical infrastructure to technical buyers, and carry a number that is already growing.