Direct answers, including the unflattering ones.

An independent technology studio and research lab that turns mathematics, artificial intelligence and data science into products — both its own and, selectively, other organisations'.

Small. The studio is new and independent, and we would rather say that than imply a headcount we do not have. It is the most useful thing to know before starting a conversation, because it determines what we can take on and how fast.

Because the method is the same in both cases. Every product starts as a modelling question with an uncertain answer, and we would rather be honest that the first stage is research than pretend the answer was obvious in advance.

We work remotely and asynchronously, in English and Spanish. Written enquiries get better answers than calls, because the reply can be thought about first.

Lluyot Labs is independent and self-funded. There are no investors to name, so we do not name any. If that changes, it will be stated here.

Problems where the answer has to be computed rather than looked up: scheduling and allocation, forecasting, dose and accumulation models, decision systems under constraints, and applied language-model work where correctness actually matters. If a spreadsheet already answers it well, you do not need us.

With a written description of the decision you are trying to make and what makes it awkward. We come back with a formulation — variables, constraints, objective, and what would count as being wrong — before anyone commits to a build. That first stage is short, and it is where we both find out whether the project is worth doing.

You do, by default, for commissioned work. Where a component is general enough to be reusable we may ask to open-source it, but that is a conversation held explicitly at the start, never an assumption.

No — that is the normal starting point, and the stage where the formulation work has the most leverage. What genuinely is too early is having neither a decision to improve nor any data about it.

Broncea, an Android application built on a dosimetric model of ultraviolet exposure, available on Google Play in six languages. It is the first public product, not the identity of the company.

An applied AI product currently in formulation, an open-source library extracted from the modelling work, and an optimization product for operational planning. We do not publish names or dates before they exist.

Because there are none yet. Listing papers we did not write would be the fastest way to lose the only asset a young lab has. The applied modelling work is real and running in production; when there is something worth submitting, it will appear with a link, a date and a venue.

We design for minimum collection first. Broncea has no backend at all — profile and history stay on the device, and location is used only to query a public forecast API. For commissioned work, data handling is agreed in writing before anything is transferred.

Not at the moment, and we do not post roles that are not funded. The careers page describes the four kinds of people the studio will need, so you can judge whether a speculative email is worth your time. They are read and answered.

Evidence of reasoning rather than a list of tools. One problem you solved properly — what made it hard, what failed, and how you knew the final answer was right — tells us more than a full CV.

There is no formal programme. Students who write with a specific technical question about the work get a real reply, and that has been the start of most useful conversations so far.