Most internships hand you a spreadsheet and a certificate. This one gives you a real problem, a measurable target, and a repository with your name on it that you can show anyone for the rest of your career.
You'll own one project end to end for twelve weeks. Not a slice of someone else's work — a whole thing, scoped against a target you have to move, built on real data, and shipped.
You'll work through the same three steps a paying client gets: scope against a measurable target before writing any model code, prototype on real data with the evaluation harness delivered alongside it, then deploy and hand over with documentation someone else can pick up. That sequence is the actual skill. Most graduates have never been made to do it.
Projects come from our own research and open-source work — vision pipelines, agent orchestration, evaluation tooling, edge deployment. Everything you build is published under your own name.
Your project ships as a public repository with you credited as the author. It is yours to show, discuss and build on afterwards. We are not going to bury it.
QVisionX is one person and a fleet of AI agents. You will not be managed through three levels of hierarchy — you work with the founder directly, weekly.
A monthly stipend for the full twelve weeks. We think unpaid technical work dressed up as opportunity is a bad deal, so we do not offer one.
At the end you get a reference describing what you actually built and what you were good at — specific enough to be useful, because it is based on real work.
We care about what you can build, not where you studied. There is no institution filter and no minimum CGPA.
Graduating this year or within the last two. Any discipline — the strongest applicants often did not study computer science.
A project, a tool, a notebook, a bot, something half-finished. It does not have to be impressive or polished. It has to be yours, and you have to be able to explain it.
You can read and write it without a tutorial open. You do not need machine-learning experience — several projects need good engineering more than they need models.
Roughly full-time. This does not work as a few hours around a full course load, and we would rather say so now than three weeks in.
No aptitude tests, no puzzle rounds, no unpaid take-home that takes a weekend.
The form below. The one question that matters is the last one — it is read properly, and it is what most decisions turn on.
Every applicant gets an answer, including a no. Being ignored after applying is the worst part of graduate hiring and it is entirely avoidable.
Shortlisted applicants have a 45-minute call about something you have built. No whiteboard algorithms, no trick questions.
With the stipend, the dates and the project you would own, in writing, before you commit to anything.
Ten minutes if you write honestly. Longer if you try to sound impressive — and it will read worse.
One cohort, five places. We reply to everyone.