We are trying to build a working model of how language models decide what to recommend — and then engineer against it. It is a research problem wearing a product's clothes. We hire slowly, deliberately, and mostly through conversations that start long before a role exists.
We would rather say that plainly than keep a page of aspirational listings up. But the team is growing, and the best hires we have made started as an email months before we posted anything.
If you have been thinking about retrieval, ranking, evaluation, or how brands surface inside generative answers — we want to read what you think, even with nothing open.
Send a short note about what you have built and what you would want to work on here. No cover letter. A link to something real is worth more than a résumé, and we reply to everyone.
Stated flatly, so you can decide whether it sounds good or awful. Both are useful answers.
Nobody has a finished answer for how to move an LLM's recommendation. That means the work is closer to research than to shipping a known feature — and it means being wrong in public, often, on the way to being right.
Our research goes out open access, with the data, under the team's name. If you want your work to be visible outside the company, that is the default here rather than a favour you have to negotiate.
There is no layer of people between you and the customer, or between you and the decision. That is the appeal and the cost — scope is broad, context-switching is real, and nothing is somebody else's job by default.
We are headquartered in Toronto and we like being in a room together for the hard parts. We work with people elsewhere when the fit is right, in overlapping hours rather than fully asynchronously.
We do not do whiteboard algorithm puzzles or take-homes that eat your weekend. We would rather look at real work and talk about it properly.
Thirty minutes. What you have built, what you are looking for, what we are actually doing. Honest on both sides about whether it is worth continuing.
We go deep on something you have already made — a repo, a paper, a system you designed. If everything you have done is confidential, we will scope a small paid exercise instead.
Two hours on a real ktau problem, with the people you would work beside. It is the closest thing we have to a trial day, and it is the part most candidates say they enjoyed.
We decide within a week of the last conversation and tell you either way. References are a genuine conversation, not a formality.
ktau is an equal-opportunity employer. We consider every qualified applicant without regard to race, colour, ancestry, place of origin, citizenship, ethnic origin, creed, sex, sexual orientation, gender identity or expression, age, marital or family status, disability, or any other protected ground. If you need an accommodation at any point in the process, tell us and we will arrange it — it has no bearing on the decision. Application details are handled under our privacy policy and are not shared outside the hiring team.
Researchers, foundation-model teams, agencies, and the merely curious — we read every email, and we tend to write back with more than we should.