Job Description
You write code that goes into systems government departments run on, and you train models that sit inside them, with a senior engineer reviewing your work. Expect to spend a good deal of your first year on data, because a system is only as good as what goes into it and government data arrives in poor condition. Three seats are open across two rural digital infrastructure programmes.
Most of what we build is not a model. Registries, credential and wallet layers, group financial records, synchronisation for intermittent connectivity, offline-first clients and the migrations that keep them alive are the bulk of the work, and they have to be engineered properly before any model on top of them is worth anything.
We do not staff a specialist for each component. Everyone here writes production software and everyone here builds models; the levels are separated by the scope a person carries rather than by the technology they work on.
Key Responsibilities
Build and maintain the components you are given: services, endpoints, jobs, pipelines and interfaces, to the standard the team sets.
Train and evaluate models on the problems you are given, and take them through review to deployment.
Build the evaluation harnesses: the test sets, the metrics, the subgroup breakdowns, and the scripts that make a result reproducible.
Reproduce a model result or a system behaviour from scratch and report where it differs.
Get data into a usable state. Ingestion, validation, feature engineering, and reconciliation against government source systems.
Write tests and keep them passing, take code review seriously, and keep the build green.
Test the way the system will actually be used, including on slow connections and older devices.
Keep documentation current: what was built, how to run it, how to retrain it, and what to check when it fails.
Ask early when something is not going to work.
Qualifications
Zero to two years of experience, or a degree in computer science, engineering, statistics or a related field with work you can show.