Software Engineer, Distributed Systems & Security

The role

You start on our decision-intelligence platform, building the private edition that runs on a customer’s own servers. After three to five months, once its design is ready, you move to our hardest research problem: a secure link between that private deployment and our public cloud platform. AI agents on both sides should reason together across it without confidential data leaving the customer’s side. A research scientist in the team designs the link and its guarantees; you build it, test it and run it.

Why this role

• Two hard problems in one role: shipping software into customer data centres, then building a research-grade secure protocol.
• You build from a researcher’s design, with the researcher working next to you.
• Security you can check: every crossing is recorded, verified and open to an outside auditor.

What you will do in your first year

• First three to five months: build the private edition of the platform for customers’ own servers, with packaging, isolation between tenants and security hardening.
• Then, once the design is ready: build the link as a typed protocol in which only permitted kinds of message cross between the private and public sides, checked on both ends.
• Integrate the policy checks on the private side that decide what may cross, within a set leakage budget.
• Record every crossing in a tamper-evident log, and build the tool an outside auditor uses to verify it.
• Build the test harness: automated debates across the link at scale, leakage measurements, tamper tests and attacks, all reproducible. Then run a reference deployment end to end.

What you bring

• A degree in computer science or a related field.
• Several years designing and owning production distributed systems in Python, Go or Rust. Show us one.
• Security engineering: authenticated messaging, strict schemas, policy enforcement, isolation between customers.
• Shipping software into customer environments: containers, Kubernetes and Helm, on-premises or offline installs.
• Test harnesses that run thousands of automated jobs and produce reproducible measurements.
• Comfort moving between product work and work from a researcher’s specification.
• You turn a vague goal into a precise work order an AI coding agent can build from, and you can show us a time you caught an agent’s code or tests being wrong.

Nice to have

LLM serving (vLLM, SGLang) and constrained decoding; cryptographic engineering (hash chains, signatures, zero-knowledge proofs); policy engines (OPA); transparency logs (Sigstore, SCITT); red-team experience.