Distinguished Systems Architect, Physical AI

Sanctuary AI•Vancouver, BC

5d•Hybrid

About The Position

Sanctuary's technical organization is built around four department Heads, each owning a vertical domain: ML, Software, Robot Software, and Hardware. This role sits as a peer to the four department Heads, reports directly to the CTO, and is accountable for ensuring the full system, across all four verticals, is scalable, interoperable and architected with a 2-3 year view toward product deployment at commercial scale. You will make architectural bets with multi-year consequences, initiate cross-cutting technical work, and be expected to disagree with existing technical directions when evidence warrants it. The right candidate will have the depth to win technical arguments on merit, the conviction to hold positions under pressure, and the judgment to know which hills are worth contesting.

Requirements

A track record of architectural decisions that shaped system-level architecture in real-world robotics or complex systems, not just individual experiments. You have made calls that constrained what others built.
Experience taking systems from R&D into production on real physical hardware, in robotics, AV, embedded, or industrial. You have shipped to the real world, not just to simulation.
AI/ML fluency to make sound architectural calls at the ML-to-runtime boundary. You do not need to be the deepest ML expert, but you do need the depth to make the call.
Experience making multi-year architectural bets and defending them under pressure, including killing directions that are not working.
Experience operating as a technical authority in a cross-functional environment: defining new architectural standards or functions, leading cross-team reviews, and influencing engineers who do not report to you.
Production Python development (3.8+) with a high bar for code quality in training and inference systems.
Working knowledge of ML frameworks (e.g. PyTorch) sufficient to reason about training and inference systems.
Working knowledge of ROS2 and real-time robotics software constraints.
Familiarity with deployment infrastructure: CI/CD pipelines, release artifact management, and build systems at scale.
Familiarity with Jira, Confluence, or equivalent engineering tooling.

Nice To Haves

Hands-on experience deploying learned models on physical robots or hardware.
A Ph.D. or advanced degree in a relevant field, or an equivalent body of shipped work.
Published or open-source work with real community adoption, as one signal of technical influence.
Familiarity with simulation environments and sim-to-real workflows is a plus (Isaac Gym, MuJoCo, or equivalent).

Responsibilities

Own the architecture of the interfaces between ML and Simulation, Software, Robot Software, and Hardware. Your decisions define how the system composes for the next 2 to 3 years.
Identify where the current architecture has structural limits (rising coordination costs, late-surfacing integration issues, hidden coupling between verticals) and build the case, with evidence, for what needs to change.
Drive architectural reviews that span vertical boundaries. Your presence in a cross-team design discussion should change its outcome.
Hold and defend the 2 to 3 year technical vision for the full platform. Form independent positions on where robot learning and simulation, deployment infrastructure, and systems architecture are headed, and work with Product to translate conviction into the roadmap.
Provide the technical lens on product go-to-market: identify where technical capability should shape what Sanctuary takes to market, and where product requirements should drive architectural change.
Present a defensible multi-year systems strategy to the CTO and leadership. You are expected to make long-horizon bets and be accountable for them.
Recommend paradigm shifts when evidence supports them. Kill architectural directions that are not working.
Own the scalability and reproducibility of the full build-test-release pipeline across verticals. The release path needs to be simple, visible, and repeatable.
Design the deployment architecture that makes Sanctuary's intelligence hardware-agnostic, deployable across in-house and third-party commercial and industrial robotic hardware.
Set architectural standards for training infrastructure at a systems level: sample efficiency, computational cost, and scalability across the full platform, not just individual experiments.
Driving system reliability across the full platform, ensuring that as we scale deployments, the systems our customers and operators depend on are robust, observable, and recoverable.
Build and improve cross-cutting infrastructure end-to-end, from simulation environment interfaces through to real-hardware deployment pipelines.
Maintain deep technical fluency across the full stack: ML training systems, sim-to-real transfer, runtime software, and deployment infrastructure. You do not need to be the deepest expert in each, but you need enough depth to make sound architectural calls at every interface.