Ruskins Academy
Learn to train and evaluate AI models.
Frontier AI systems are taught by people with real expertise in a field. If you have that depth, your expertise is the qualification — what is missing is the craft of turning it into training and evaluation work. That is what we teach.
The work
What this work actually is.
A model that can pass a general knowledge test still fails at the things a practitioner does daily. Closing that gap takes people who know what good looks like in a specific field, working through problems with the model and saying precisely where it went wrong.
In practice that means designing a realistic problem from your own work, running a structured conversation with the model, comparing the responses it produces, choosing the better one against explicit criteria, and writing down why. The judgement is the product.
It is remote, project-based work, and demand for it exists across a range of organisations that build and evaluate AI systems. We prepare you for that category of work rather than for any one place that offers it.
Who it suits
Depth in a field, whichever field that is.
No AI background is expected and none is required. What matters is that you can look at work in your domain and say exactly why it falls short.
- Finance
- Analysts, accountants, actuaries, investment and FP&A professionals
- Healthcare
- Clinicians, pharmacists, researchers and allied health professionals
- Law & policy
- Practising lawyers, compliance specialists and policy analysts
- Engineering & STEM
- Engineers, scientists and technical specialists
- Operations & strategy
- Consultants, operators, project and supply chain professionals
- Education
- Teachers, lecturers, curriculum designers and subject specialists
What you learn
Six modules, each building on the last.
- 01
Task design
Writing realistic problems from your own field — the kind that would take a competent professional hours of genuine work. Most weak training data fails here, in scenarios too thin or too artificial to reveal anything about a model.
- 02
Structured conversation
Running multi-turn exchanges with a model deliberately rather than conversationally: establishing the task, probing where you suspect weakness, and steering without leading it to the answer you want.
- 03
Comparative evaluation
Judging two candidate responses against explicit quality dimensions and defending the choice. This is the core skill, and the one that separates useful evaluators from fast ones.
- 04
Writing feedback
Turning a judgement into feedback specific enough to act on. Naming what is wrong, why it is wrong in your domain, and what correct would have looked like.
- 05
Documenting failure modes
Recognising when a model is confidently wrong in ways a non-expert would miss, and recording the pattern rather than the single instance.
- 06
Domain-grade deliverables
Producing the documents, models and analyses this work runs on, to the standard your field expects rather than to the standard a demo tolerates.
How it runs
One to one, remote, and built on your own work.
Taught directly
You work with an instructor rather than through recorded material. Sessions are shaped around your field and what you are finding difficult.
Practice first
You produce real evaluation work from early on. Every module ends with something you made, reviewed line by line.
A portfolio, not a certificate
You finish holding evidence of the work you can do — the artifact that actually matters when someone is assessing you.
Questions
Before you apply.
Do I need a technical or AI background?
No. The scarce thing in this work is genuine depth in a field, which you already have. Everything specific to AI is taught here, and none of it requires programming.
What field do I need to be in?
Any field where you have real, senior-level judgement — where you could look at a piece of work and say precisely why it is wrong. The domain matters far less than the depth.
Will this get me a job?
We train you in the craft and you finish with a body of work that demonstrates it. We do not place people, and we make no promises about hiring — anyone who guarantees you a role in this market is not being straight with you.
What do I finish with?
A portfolio of completed evaluation work in your own domain: task scenarios you designed, model exchanges you ran, comparative judgements you made and the written reasoning behind each.
How much of this is theory?
Very little. You work on real tasks from your own field from early on, and the feedback you get is on work you actually produced.
Apply to the programme.
Tell us what field you work in and how long you have worked in it. We will come back to you about whether the programme fits, and what it would involve.