AI & Machine Learning Company in Perth

Perth AI budgets track how close the model gets to the pit. A scoped AI proof of concept at Perth rates stays in the office. A custom production ML model — predictive maintenance on a haul truck fleet, ore grade estimation from camera feeds, LNG compressor anomaly scoring — gets closer. Full production AI systems integrated with OSIsoft PI, autonomous haulage, or Pawsey HPC are on site. Scope drivers include data audit, baseline model, and a hosted demo on a representative dataset. Perth rates sit above Adelaide and Brisbane because of the mining talent premium and the AWST timezone scarcity (most Australian AI talent sits in Sydney and Melbourne). Codazz quotes fixed-fee on agreed scope rather than open time and materials, with progress payments tied to acceptance milestones.

Yes. The Pawsey Supercomputing Centre, jointly funded by CSIRO, Curtin, Edith Cowan, Murdoch, and UWA, runs Setonix — currently Australia's most powerful research supercomputer — alongside the Nimbus research cloud. Access is merit-based through the Pawsey allocation rounds. We help WA clients write competitive proposals for geophysical inversion, multi-physics simulation, large-scale climate and cyclone modelling, and foundation model pre-training where commercial cloud GPU spend would exceed AUD $200,000 per quarter. For industry clients without academic affiliation, the Pawsey-on-Demand commercial channel is available. We have built workflows that hand off pre-processing on AWS, training on Setonix or Nimbus, and inference back on commercial cloud for production serving, with full lineage tracking through MLflow.

The Security of Critical Infrastructure Act 2018, amended by the SLACIP Act in 2022, declared mining (specified critical minerals), oil, gas, and electricity assets as critical infrastructure. Responsible entities — including Rio Tinto, BHP, Fortescue, Woodside, and Chevron Australia — must register assets, implement a Critical Infrastructure Risk Management Program (CIRMP), report cyber incidents to the Australian Cyber Security Centre within twelve hours (significant) or seventy-two hours (relevant), and may receive Government Assistance Measures directions. Any AI system that controls, monitors, or materially influences mining or energy operations falls inside that perimeter. Our SOCI-aware builds ship with an asset register update, a CIRMP-aligned model risk assessment, OT-IT network segmentation that the ASD Essential Eight reviewers accept, and an incident response playbook that hits the twelve-hour notification window.

We build supervisory AI and ML around autonomous mining stacks rather than replacing them. Rio Tinto's Autonomous Haulage System runs on Komatsu FrontRunner; BHP and Fortescue use Caterpillar Command and integrated platforms. Our scope is the layer above — fleet health analytics combining payload, fuel burn, tyre temperature and vibration to predict failures forty-eight to ninety-six hours out, anomaly detection on drill and blast data, computer vision on shovel and conveyor cameras for spillage and tramp metal, and reinforcement learning for dispatch optimisation. We integrate with PI System historians, Modular Mining DISPATCH, and Komatsu KOMTRAX, and we explicitly do not touch safety-rated control loops — those stay with the OEM. Models ship with explainability, drift monitoring, and a human-in-the-loop review queue for the Perth operations centre.

Yes. Western Australia hosts the world's largest hard-rock lithium operations (Greenbushes via Talison, Pilgangoora by Pilbara Minerals, Mt Marion by Mineral Resources and Albemarle), expanding rare earth plays (Lynas Mt Weld, Iluka Eneabba refinery), and a deepening critical minerals pipeline. We build exploration targeting models on integrated geophysics (gravity, magnetics, seismic), geochemistry, drill assay databases, and structural geology from Geological Survey of Western Australia open data plus client proprietary datasets. Approaches include gradient boosting on engineered geological features, convolutional networks on geophysical grids, and self-supervised learning on unlabelled drill core imagery. Outputs are mineral prospectivity maps with calibrated probabilities and SHAP attributions that geologists can defend in joint venture and ASX disclosure contexts. We do not replace exploration geologists — we sharpen their target ranking.

We default to AWS ap-southeast-2 (Sydney) for storage, training, and inference, with ap-southeast-4 (Melbourne) as DR. GCP australia-southeast1 and australia-southeast2 are equivalents when clients prefer Google. Microsoft has no Perth Azure region, so Azure-anchored clients run from Australia East (Sydney) with ExpressRoute back to Perth carrier hotels at Equinix PE1/PE2 or NextDC P1. For mine sites in the Pilbara with intermittent connectivity (Telstra, Optus, and Starlink), we run edge inference on NVIDIA Jetson or AWS Snowball Edge, syncing model updates and telemetry on schedule. All residency decisions are documented in the data processing record under Australian Privacy Principle 8 (cross-border disclosure), and we will not move data offshore without an Australian Privacy Principle 8.2 exemption or contractual safeguards your privacy officer signs off on.

A typical predictive maintenance model on a haul truck or processing plant fleet takes sixteen to twenty-six weeks from kickoff to production. Weeks 1 to 4 cover historian extraction (PI System, AVEVA), data quality assessment, and SOCI risk classification. Weeks 5 to 14 are feature engineering, modelling, iteration, and shadow-mode evaluation against the maintenance team's existing playbooks. Weeks 15 to 22 are MLOps, drift monitoring setup, OT network segmentation review with the client's cyber team, and acceptance against agreed sensitivity and precision targets. Weeks 23 onward is gradual rollout with the operations centre. If telemetry is sparse or labelling is needed (failure modes, ground truth), add eight to twelve weeks. We have shipped to this cadence inside Pilbara operations and Perth-based control rooms.

When a project requires genuine research — novel geophysical inversion architectures, multi-physics simulation surrogates, rare-event detection on imbalanced safety datasets — we scope collaborations with UWA's Data Institute, Curtin Institute for Computation, or Curtin's Centre for Exploration Targeting rather than overselling in-house capability. UWA's Australian Institute for Machine Learning satellite presence and Curtin's top-ranked mining engineering program give Perth genuine depth on resources AI. Our core team handles applied engineering, MLOps, and productionisation, which is where most Perth AI projects actually stall. For standard work (RAG, fine-tuning, classical ML on tabular telemetry, computer vision on known architectures) no academic partner is needed. We will tell you up front which bucket your problem fits into.