Where New Zealand AI stands
The country is small, and its artificial intelligence research is concentrated in a handful of universities, public applied-research institutes and technology companies, notably in agriculture, health, geospatial work and digital services. There is no large market for AI internships comparable to that of some more populous countries: opportunities tend to arise from joint projects between a laboratory and an industry partner.
This changes the way you search. You do not approach a recruitment department first, but a team, a lecturer or a summer programme. A student from abroad therefore benefits from identifying researchers whose work matches their level, reading their recent publications and proposing a specific contribution.
The site is informational: it does not place anyone and guarantees neither a position nor funding.
The university summer scholarship model
The Artificial Intelligence Institute at the University of Waikato describes a highly structured pathway for undergraduates: a summer research scholarship, with a full-time project of about ten weeks during the November-to-February term, carried out with an industry partner. According to the page consulted in 2026, the partner funds the scholarship, the university co-funds a limited number each year and a public innovation agency may add a supplement. At the end, the student hands in a research poster.
The announced calendar follows a precise rhythm: topics are proposed around July or August, applications and interviews take place in September, contracts are signed in October and the work starts in November. For an applicant from abroad, this rhythm means making enquiries in the spring, well before the official opening.
Note that this scheme assumes you are enrolled in a paper, that is, a course carrying credits, at the institution. A university exchange can therefore act as a way in.
Public institutes, innovation and companies
Outside the universities, public applied-research bodies host students in the summer. The Bioeconomy Science Institute, which merged several former science institutes devoted to agriculture, food and natural resources, publishes a page dedicated to its internships. The profiles sought there vary by project: statistics, programming, image processing, modelling. Each topic has to be read carefully, because not all are open to the same levels.
The agency Callaghan Innovation, which supports business research and development, acts mainly as a funder or facilitator of partnerships. It does not recruit interns on demand, but its website helps you understand how a university project meets an industry partner.
A few private software, health and agritech companies also take on data science interns. Their offers are posted on their careers pages, often late in the calendar year for a southern summer.
Visa, placement agreement and the intern’s status
Before imagining the topic, check the right to work. A student enrolled at a New Zealand institution holds a student visa whose employment options are restricted; a placement built into the course may follow specific rules. A young person from some countries can also look at a working holiday visa, whose age limit, duration and restrictions are set by Immigration New Zealand and may change.
A scholarship is not an employment contract, and an internship is not always a job: the statuses differ, and so do social protection, tax and insurance. Ask in writing whether you are paid, by whom, and which document governs the assignment.
Data protection matters too. Projects that handle personal information fall under New Zealand privacy law, and data belonging to Māori communities is the subject of particular thinking about how it is governed. An intern must never take away or reuse an employer’s data without permission.
A document-classification prototype, step by step
For an application, a modest but well-measured project is worth more than a vague ambition. Document classification is a good testing ground, because it forces you to describe the whole chain.
- Gather a fictional or copyright-free corpus: invented invoices, letters and forms.
- Scan the documents, or simulate scans of differing quality.
- Extract the text with an optical character recognition tool.
- Classify each item with a simple model, then a richer one.
- Measure the results on a test set kept aside.
- Plan a human review for uncertain cases.
The most interesting point is separating the errors. An invoice that is filed wrongly may come from a blurry scan, a faulty extraction or a model that is too weak. Documenting these three causes shows a genuinely scientific approach, far more convincing than a single score.
A worked example: reading an error rate without being misled
Imagine, with round and purely hypothetical figures, a test on 500 fictional documents split into five categories. The prototype classifies 450 correctly, which is 90%. That sounds good, but you need to look further.
Suppose that 40 of the 50 errors involve low-quality scans, and only 10 come from the model. The conclusion is no longer the same: improving the scanning would pay off more than changing the algorithm. If, in addition, 300 of the 500 documents belong to a single category, a system that always answered with that category would already reach 60%.
Presenting these nuances, with a simple confusion matrix and a few examples of failures, proves that you understand what an indicator measures. A supervisor will read this rigour as a sign of maturity, even if your score is modest.
Skills to work on before applying
Technical foundations
- programming in Python and use of machine learning libraries;
- statistics, probability and linear algebra at undergraduate level;
- version control and reproducibility of experiments;
- cleaning and documenting data.
Project skills
- formulating a testable question and an evaluation protocol;
- writing a short report in English, with readable figures;
- presenting a result to a non-specialist audience, which a poster format demands;
- discussing limits, biases and inappropriate uses.
A small public repository, tidy and well commented, weighs more than a list of keywords. It is enough that it contains instructions for re-running the experiment.
Writing to a researcher or a team
An effective message fits in a few lines. Describe your course, cite a paper or a project from the team that you have genuinely read, explain what you can contribute and give your dates of availability. Attach a one- or two-page CV and the link to your code repository, if you have one.
Avoid mass mailings. Teams are small and they recognise each other’s inboxes. Also mention your administrative situation: home institution, whether a placement agreement exists, planned funding. Finally, ask one precise question, for example about the level expected or whether a supervisor could review your work.
If no reply arrives after two weeks, a short follow-up is acceptable, once only.
Common mistakes and fallback options
The most frequent difficulties are well known. Some applicants promise results that their protocol cannot establish. Others neglect the visa question or discover too late that the scholarship requires local enrolment. Others still confuse an internship, a doctorate in partnership with industry and a simple course project.
If the calendar does not suit you, several fallback options exist: a remote project with a laboratory, a contribution to open-source software, an internship in a laboratory at home or in Europe linked to a New Zealand team, or an exchange semester so that you can come back and apply locally. In every case, confirm the exact conditions with the institution and the team before committing.
Frequently asked questions
Are there AI internships open to foreign students?
Yes, but they are few and often go through a university, a research institute or a summer programme. A university exchange or a placement agreement with your school makes access easier. Contact teams early, stating your level, your skills and your dates of availability.
Is the summer research scholarship a salary?
Not necessarily. A scholarship is funding tied to a project and to student status, whereas a salary rests on an employment contract. Amounts and conditions vary with the year and the partner. Ask for the detail in writing, along with the insurance and tax rules that apply.
When should you get in touch?
The rhythm of a summer project starts several months before the departure: topic proposals in the middle of the year, applications and interviews in early southern spring, then contracts in the autumn. For a November start, it is best to make contact between the European spring and summer.
Do you have to be enrolled in New Zealand to apply?
For some summer scholarships, yes, because they require enrolment in a course at the institution. Other internships welcome students from elsewhere. Read the conditions of each scheme, and check with the team whether an exchange student can apply.
What level of programming is expected?
A good level in Python and the ability to use common machine learning libraries are enough for many undergraduate projects. More important still, you need to know how to evaluate a model, document your trials and explain its limits. A small commented code repository usefully supports an application.
Can you use real data in an application?
It is better to avoid it. Use fictional, synthetic or openly licensed data, especially if personal information could appear. Privacy rules are taken seriously, and an employer will appreciate that you respect them from your very first prototype.