hotloop.ai
Let's chat
Back to hotloop.ai
The research

Twelve months, under real compliance risk.

The method we sell was not designed in a workshop. It came out of New Zealand's largest study on AI in vocational education, where getting it wrong had consequences.

What we were asked to do.

ConCOVE Tūhura, the Construction and Infrastructure Centre of Vocational Excellence, asked whether generative AI could produce assessment material at scale without dropping below the standard a moderator would sign.

That is a harder question than it sounds. Assessment carries legal weight. Material that fails moderation is not an inconvenience, it is a compliance event, and the people affected are learners whose qualifications depend on it.

Karl was principal investigator. The programme ran for twelve months under ethics review and data sovereignty rules, and produced an ethical framework written in response to Te Tiriti o Waitangi.

A reviewer at a desk checking drafted work against a standard before signing it off.
What it produced

The numbers we can stand behind.

1,000+assessments generated

Across the twelve-month programme.

10,000+questions reviewed

Every one checked against the standard.

135unit standards modernised

Delivered with Te Pūkenga and Competenz.

Figures from the ConCOVE Tūhura programme. Data sovereignty was held throughout, which was a condition of the work rather than a feature of it.

What we learned

Three findings that shaped the method.

FINDING 01

The model was never the hard part.

Generating material at volume was straightforward within weeks. Everything difficult was downstream: deciding what counted as good, who was allowed to say so, and how that decision got recorded.

FINDING 02

Cheap checks beat expensive review.

Mechanical checks run before an expert looks removed most of the failures for almost no cost. Expert attention is the scarce resource, and spending it on obvious faults is the waste nobody counts.

FINDING 03

Governance had to be structural.

Governance written as policy was ignored under deadline. Governance produced by the workflow itself survived, because there was no faster path that skipped it.

Delivered work

Where it has been used since.

Delivered by Karl across hotloop and Epic Learning engagements.

RESEARCH
New Zealand's largest study on AI in vocational education

Principal investigator on the ConCOVE Tūhura programme. Over 1,000 assessments generated and 10,000 questions reviewed, with data sovereignty held throughout.

LIVE
135 unit standards, modernised

Mechanical Engineering unit standards modernised for Te Pūkenga and Competenz, with digital assessment deployed.

LIVE
NZF Learn, 4,000+ active users

The learning platform for New Zealand Football, with 4,000+ active users.

LIVE
AI Activator, with Callaghan Innovation

Module development for Callaghan Innovation's AI Activator programme.

LIVE
Assessment against state legislation, Australia

An AI assessment engine for the construction sector, built to overlay state legislation.

LIVE
Print industry, end to end

Assessments, unit standards and workbooks delivered from standard through to learner-facing workbook.

Start with one workflow

Bring the job that eats your week.

We open with a discovery session, map it with your people, and set the timeline from what we find.