About the Job
Founded in Australia in 2019, our mission is to create the most efficient AI infrastructure by combining cutting-edge technology with a steadfast commitment to sustainability.
At Firmus, we are unique in our approach. We design, build, and operate a new class of digital infrastructure – the AI Factory. Through our model-to-grid technology approach, we have pushed the boundaries of multi-generational liquid cooling systems, energy management, AI software orchestration, and construction. Fo...
Our large-scale GPU cloud platform, Firmus AI Cloud, is purpose-built to deliver energy-efficient AI compute at scale to customers.
Key Responsibilities
Scientific machine learning and surrogate models
Develop and qualify surrogate models for bounded thermal, airflow, cooling and infrastructure scenarios. Start with a well-defined simulation-backed use case, selecting convolutional models, neural operators or graph-based approaches according to the data representation, fidelity and latency required. Extend to geometry-generalising approaches only when data coverage and validation justify them.
Engineering data and evidence
Build reproducible pipelines from Ansys Fluent and other engineering tools into model-ready datasets. Preserve geometry, mesh, equipment, boundary-condition, units and solver-version lineage, and define canonical field and graph representations with the relevant engineering owners in Engineering and Technology. Design evaluation splits that prevent leakage across related cases and test unseen operating conditions and configurations. Treat approved solver studies and measured evidence as reference inputs, not as data to be stripped of engineering context.
Production services and integration
Deliver tested, versioned software rather than isolated notebooks. Build training pipelines, model registry and qualification workflows, inference APIs and scenario-orchestration integrations. Own model packaging, qualification workflows, automated testing, monitoring and technical documentation in partnership with platform engineering. Deploy surrogate model endpoints through the AI & Applications inferencing platform and execute on shared GPU capacity managed by the AI & Applications scheduler. Connect model outputs to the engineering optimiser, evidence-linked design records and visualisation services, including Omniverse and OpenUSD where appropriate.
Qualification, uncertainty and safe use
Agree acceptance criteria with engineering owners before a model is released. Evaluate field accuracy, critical local errors, physical consistency, inference latency and performance against simpler baselines. Implement applicability-domain checks, calibrated uncertainty where supported, out-of-distribution detection and clear rejection or solver-fallback behaviour. Make every result traceable to its inputs, model version, validation evidence and limitations. Keep operational evaluation in shadow mode unless separately approved; surrogate services must not bypass independent controls or protection systems.
Whole-factory decision support
Integrate surrogates into scenario screening, design comparisons, commissioning rehearsal and operating-envelope studies. Work with engineering teams to assess delivered compute performance, whole-factory energy efficiency, auxiliary loads, cost and risk within explicit system boundaries. Introduce workload-intent and model-to-grid scenarios as the evidence base matures. Prioritise improvements that shorten engineering decision cycles without weakening assurance, and build reusable capability across factory configurations rather than one-off demonstrations.
Required Skills & Abilities
You bring deep, hands-on experience in scientific machine learning, computational engineering or simulation-led software, with a track record of taking models from research into dependable production services across multiple projects. You have technically led at least one cross-disciplinary programme spanning ML, simulation and software delivery. You are strong in Python and PyTorch, and can develop, train, profile, evaluate and deploy models on GPU infrastructure. Your experience includes surrogate modelling, neural operators, graph neural networks, reduced-order modelling, physics-informed learning or closely related methods.
You are comfortable working with simulation data, meshes, geometry, boundary conditions and numerical validation. You understand how discretisation, convergence, conservation, data coverage and measurement uncertainty affect a learned approximation. You can distinguish interpolation from unsupported extrapolation and explain when a surrogate should defer to an authoritative engineering calculation.
You have built maintainable software with APIs, containers, version control, automated tests, CI/CD and observability. You can design reproducible data and model contracts, diagnose failure modes and work constructively across engineering disciplines. A degree in engineering, computer science, applied mathematics, computational physics or a related discipline, or equivalent practical depth, provides a relevant foundation.
Experience with NVIDIA PhysicsNeMo or Modulus, PyTorch Geometric, graph or operator learning, uncertainty quantification and multi-GPU training is highly regarded. Familiarity with Ansys Fluent or PyFluent, OpenFOAM, ETAP or comparable engineering tools, together with CAD-to-simulation workflows, OpenUSD, Omniverse, Kubernetes or on-premises accelerated computing, would be valuable. Data-centre thermal systems, liquid cooling, electrical infrastructure or GPU/HPC experience is useful; practical scientific-ML and software-delivery evidence matters more than familiarity with every tool. Experience operating as a technical anchor across simulation, ML and software teams — setting standards, reviewing designs and mentoring engineers is expected at this level.
