§ 00 About
Academic habits, pointed at real problems
Prosop.ai is a small, independent AI consultancy. We're the people you call when you want the truth about whether something will work, not a proposal for building it regardless.
Where this came from
A research training leaves you with a few stubborn habits: define the question before you pick the method, write down in advance what would prove you wrong, and treat a result you like with exactly as much suspicion as one you don't. Those habits are deeply unfashionable in consulting, where confidence sells better than calibration. They also happen to be the difference between an AI project that works and one that merely demos.
So we kept them. We will tell you when the data isn't there. We will show you the error bars. And if the honest answer is that you don't need machine learning for this, you'll hear it from us early, while it's still cheap.
Who you'll actually work with
Debora Andrade — PhD in Physics. Worked on non-linear optics at Oxford and Imperial College London. Happiest when asked to compute the expected value of a life-changing decision, and quite incapable of doing it briefly.
Small by choice. You get the person you met in the first meeting, for the whole engagement, rather than a pitch team followed by whoever was on the bench.
§ 01 Principles
Three things we don't negotiate
Security by design
GDPR-native from the first sketch, not bolted on before launch. We assume your data is a liability as well as an asset, because it is.
Honest guidance
We're not sellers, we're problem-solvers. You get the uncertainty along with the estimate, and a straight answer when the straight answer is no.
True partnership
Coupled oscillators exchange energy until they move together. Same idea here: your mission becomes ours, from first sketch to long after deployment.
ask us about the error bars. please.
§ 02 Open position
Wanted: one stubbornly curious person
We're hiring a junior colleague. Straight out of university is ideal. Arriving from somewhere else entirely is equally ideal. You do not need to arrive already building language-model apps. We can teach the tools. We are selecting for how you think.
What you'd actually do
You would write software, not only advise. You'd work on real client problems from the first week, alongside the founder, not on a training track. That means learning how we design and build machine learning systems — including language models, when they are actually the right tool — and then doing the unglamorous parts: data, evaluation, the work that decides whether something ships. The systems are small and real: they go in front of clients, they get evaluated, and they have to keep working.
You'd also be in the room while we advise clients on what is worth doing, what it will cost, and where the sharp edges are. Strategy is not a junior role. Being present while it happens is how people grow into it.
You might be a fit if
- You are meticulous to a degree that occasionally irritates people, and you've made peace with that
- You would rather say "I don't know" in front of a client than discover later that you were bluffing
- Given an easy problem and a hard one, you're already thinking about the hard one
- You get genuinely interested in industries you'd never have chosen for yourself, whether that's logistics, insurance or agronomy
- You like people. Much of this work is real conversations with experts in their own field, and you'd find that a perk rather than a tax
- You make things up in the good sense: unusual approaches, odd analogies, the occasional diagram nobody asked for
A computer science degree would be a genuine advantage. So would physics, maths or another quantitative course. None of them is a gate. You do not need to arrive already having shipped language-model apps.
What we are not asking for
- Years of experience. This is a first job and we mean it
- A long list of frameworks you've touched. We'd rather see one thing you understood properly
- A flawless, professional-sounding cover letter. We would much rather read something that sounds like you
Remote, but not adrift
You'd work from home, anywhere in the UK, with occasional trips to London to meet clients face to face. Junior and remote can be a lonely combination, so it won't be one here: you'll work directly alongside the founder on real client problems from the first week, rather than being parked on a training track until you're deemed ready. Laptop and tools provided. London trips are booked and paid for; they do not come out of your salary.
How to apply
Send an email. Instead of a cover letter, tell us about a problem you found genuinely hard: what you tried, what didn't work, and how you knew when you were wrong. Half a page is plenty. A CV is welcome, but it's the smaller half of the decision.
Use whatever tools help you write it. We build with language models for a living, so it would be strange to object. But we are reading for your reasoning, and the gap between your thinking and a model's impression of it is wider than people expect.
null results welcome here too
Come with a problem, not a brief
The best conversations start with something that's genuinely annoying you, rather than a solution you've already picked.