Machine learning development in Perth
Processing plants in Western Australia, for iron ore, gold, nickel and lithium, record thousands of readings a minute and still lose throughput to problems that build slowly: a gradual drift in a circuit, a feed change nobody flagged, a pump that is starting to fail. The metallurgists usually know what to look for. They do not have time to look at everything.
We build models on historian data read out one way from the plant network: anomaly detection that flags an unusual pattern early, forecasts of throughput and recovery that planners can use, and explanations that show which readings drove each alert. The model advises; your metallurgists and operators decide. On FM360 we took repeat-fault clustering from experiment to production, and the lesson holds for plants: people act on an alert only when they can see why it fired.
Perth runs mines it cannot see. Rio Tinto controls trains, trucks and drills in the Pilbara from its operations centre in Perth, BHP runs much of its Western Australian iron ore logistics from a remote operations centre in the city, and Woodside, Fortescue and Chevron Australia keep their head offices here. Around them sits a large sector of engineering consultancies, contractors and mining equipment, technology and services firms. Most of the software work in the city exists to serve that industry, and it comes with that industry's habits: safety first, careful change control and very little tolerance for downtime.
The line that matters is between operational technology and everything else. Control systems, autonomous haulage and plant automation belong to specialists. The opportunity for outside engineers is on the business side of that line: maintenance and shutdown planning, historian data read out for analysis, contractor and workforce systems for a fly-in fly-out workforce, reporting, and the internal tools that stitch them together. Energy and utility businesses need the same kind of work, and beyond resources the state has its own buyers, from agencies and universities in the city to the wineries of Margaret River selling across the continent.
Hiring is shaped by the resources sector, which competes hard for experienced engineers and can pay for them. Indeed puts the average base salary for a software engineer in Perth at A$100,253, from 59 reported salaries, and a consultancy or scale-up bidding against a mining major for the same person often waits a long time. That is where we fit: senior engineers on a defined piece of work within days, with the code in your repositories from the first commit.
Our four hours of daily overlap land in your afternoon, moved to suit your team. Perth is the closest Australian capital to India's time zone, and neither Western Australia nor India uses daylight saving, so the window sits at the same point in your day all year. Your morning is for your own team and the site calls; the afternoon is for stand-ups, reviews and decisions with the engineer who writes your code.
We are the right fit for defined work on the business side of the operation: data, reporting, maintenance and workforce tools, and contractor systems. Senior engineers start within days, the first piece is a fixed-price two weeks, and we can carry the roadmap through the months before your own hire starts. Agencies can put the work under their own name.
Four ways this arrives.
Someone got good results on a laptop and nobody can reproduce them. We turn it into a training pipeline with pinned data and versioned artefacts, put a serving endpoint in front of it, and record what it was evaluated on so the next version can be compared.
Hundreds of conditions routing tickets or flagging risky orders, each added after an incident. We measure what the rules get wrong on your labelled history, then test whether a classifier beats them. Sometimes the answer is fewer, better rules, and we say so.
Demand, load or staffing estimates that miss often enough to be ignored. We start with a plain seasonal baseline, backtest against the periods you actually plan around, and add complexity only where it beats that baseline on those periods.
Duplicate records, recurring faults, near-identical products under different names. Embeddings group text by meaning where keyword matching cannot, and we build the screen that lets a person correct a group rather than trust it blindly.
Repeat-fault detection that began as UMAP and HDBSCAN over 33,000 work orders and shipped as nightly incremental clustering on the same embedding model.
Read the write-up →Asked by Perth teams.
How do you work with teams in Perth?+
We are in Bengaluru, and four hours of every working day overlap with your afternoon, moved to suit your team. Perth is the closest Australian capital to India's time, and neither side changes its clocks, so the window stays put all year. Stand-ups, code reviews and design calls happen live in it. We work in your tools, whether that is Slack, GitHub or Linear, and every call is with the engineer who writes the code.
Do you build for mining and energy companies?+
Yes, on the business side of the operation. We build pipelines that read historian data out for analysis, maintenance and shutdown planning tools, contractor management platforms, apps that put a FIFO worker's roster, flights and documents in one place, and dense planning views for remote operations teams. We integrate with the maintenance, HR and procurement systems you already run, and start with the problem that costs you most on site.
How does your rate compare to hiring in Perth?+
Indeed puts the average base salary for a software engineer in Perth at around A$100,000, and the resources sector often pays more to win experienced people. Our published rate is $35/hour with a $5,000 minimum, or $5,400 a month for one embedded engineer. There is no recruiting time and no employment overhead, and work can start within days. The first piece is a fixed-price two-week engagement at $2,800, so you judge us on what ships.
What do the first two weeks look like?+
You share the context, with an NDA signed before anything specific. Within two days you are on a 30-minute call with the engineer who would do the work, and within a week you have a written plan. Then comes a fixed-price two-week piece for $2,800: one real thing, such as a maintenance report, a contractor portal screen or a data feed, built in your repositories and reviewed with you in your afternoons.
Do we need machine learning, or is a rules engine enough?+
Often a rules engine is enough, and it is cheaper to run, explain and audit. If the people who make the decision today can write it down, write it down. A model earns its place when the rules keep multiplying, the inputs are messy text or images, or the pattern shifts faster than anyone can edit the rules. We check against your history before recommending either.
Why not just call a hosted model API?+
Sometimes you should. With low volume, messy language and no labelled data, a hosted model and a good prompt ship quickly. A trained model wins when per-call pricing starts to hurt at your volume, when latency matters, when data cannot leave your systems, or when you have labelled examples a small model can learn from. We price both paths before you commit.
How much data do we need?+
For classification, usually fewer labelled examples than people fear, and embeddings can group text with no labels at all. Forecasting is the opposite: you need enough history to cover the cycles you plan around, and a single year says little about a yearly pattern. We give you an answer after looking at your data, not before.
What happens when the model gets worse?+
It will, because the world it learned from changes. We log inputs and predictions, compare them with outcomes as they arrive, and alert when accuracy or the input mix drifts past a threshold you agreed. Retraining is a scripted pipeline with an evaluation gate, so a new model replaces the old one only when it does better on the same test set.
Can you build a recommendation system?+
Yes, and we would start without a learned model. Co-occurrence and embedding similarity, served from Postgres or a vector index, beat a most-popular list for most catalogues and are easy to debug. A learned ranker comes after that, once you have enough interaction data to train and evaluate one. A learned ranker is something we build from our work with embeddings and clustering, starting from a simple baseline you can measure it against.