Skip to content
TYPE8

AI & ML engineering

An AI system without an evaluation harness is a demo. We treat evaluation, drift monitoring and human oversight as first-class deliverables, not a later phase.

A$4.5BPROGRAMME · PENDING VERIFICATION · C-001

Launch model

Core

Boundary

We do not present autonomous decisioning without a human control point.

Accountable principal

Sim Cuthbert

When to engage

Four observable triggers.

  • 01A copilot pilot cannot pass security or privacy review.
  • 02Model performance is asserted rather than measured.
  • 03There is no inventory of models running in the business.
  • 04Retrieval quality is unknown and unmonitored.

What we deliver

Modules, artefacts and ownership.

Client inputs, dependencies and the operating owner are agreed before the first sprint, not discovered during it.

  • GenAI, RAG and agent engineering
  • Evaluation harnesses and benchmarks
  • MLOps, deployment and reproducibility
  • Model monitoring and drift control
  • Human oversight and escalation design

Proof

Programme intelligence for a national broadband rollout

A programme data platform with an explicit KPI contract, reconciled site status model, governed pipelines and an executive intelligence layer with alerting.

Delivered by our principal while at a national delivery organisation.

Read the full outcome

Risk and assurance

Security, privacy, AI assurance, accessibility, data quality and human oversight are tested as acceptance criteria on this practice — not appended as a closing report.

Standards
WCAG 2.2 AA, Essential Eight and ISM mapping where in scope.
Data residency
Australian by default; offshore processing named and approved in writing.
Support boundary
We do not present autonomous decisioning without a human control point.
Response standard
Principal response within one business day, AWST.
Open the Trust Centre