Ur AI selects 10 interns from 312 applicants in METI's Global South internship program

Ur AI is hosting its first intern cohort through the Global South - Japan Tech Talent Internship (GS-JTI), a program run by Japan's Ministry of Economy, Trade and Industry (METI) that connects students from Global South countries with companies in Japan.

We received 312 applications for our Document Intelligence and VLM R&D internship, and selected 10 students. They join us in December, and we are excited to welcome them.

The cohort

• 10 students from top universities such as the IITs and BITS Pilani in India, and BUET in Bangladesh.
• Four weeks of R&D between 7 December 2026 and 15 January 2027: two weeks remote, and two weeks with the team in Tokyo from 21 December to 1 January.
• All 10 chose to come to Tokyo.

How we selected: a test of AI-native ability

Alongside the program's own shortlisting and interviews, we ran an assessment of our own, built to measure AI-native ability. Candidates worked on a real document problem and could use any AI tool they liked. We were not testing whether someone can work without AI. We wanted to see how they work with it: how they frame a fuzzy problem, what they hand to the model, where they check its output, and whether they notice when it is confidently wrong.

What they will work on

The interns will work on a mix of projects from our R&D roadmap.

• Nebula projects: post-training vision-language models for document conversion, and language models for retrieval and reasoning, the work behind Nebula and Nebula Sovereign, which runs on open-weight models we post-train ourselves.
• A first step into science: we will start working on one of the Millennium Prize Problems, for example the Navier–Stokes equations or one of the open problems in pure mathematics, with post-trained models. We will pick the problem with the team. The aim is to build our research muscle on a problem far harder than any benchmark, and to show that our methods carry over beyond documents.

We are glad to have more talented people pushing on these research fronts with us.

Why we are betting on AI-native talent

I believe the people who grow up with AI as a given will build things the rest of us would not think of. For today's students, an AI model is not a new tool to adopt. It is part of how they learn, write and solve problems, at an age when ways of thinking are still forming. They have nothing to unlearn.

Those of us who built our careers before AI approach it differently. We re-think how we work, and we are very good at it: making today's workflows faster, cheaper and more reliable. That is real value, and much of what AI delivers right now.

But the next generation of products and services will not be today's workflows with AI added on. They will be designed around AI from the start, and I expect many of them to come from people who never knew another way of working. Putting that instinct next to the experience in our team is what this program is about for us.

Thank you

Thank you to METI and the GS-JTI program office for making this possible, and to all 312 applicants for the time and care you put into the problem. If you are a student who wants to work on hard document AI problems, follow our updates at ur-ai.net/blog, or write to us at contact@ur-ai.net.