Find the clinical risks. Improve the AI.
For health-tech teams building medicines information, prescribing support and clinical AI. I help you evaluate responses, investigate failures and design testable improvements. Led by Rob Rumney, pharmacist, independent prescriber and SCRUB-in co-founder.
- Clinical
- Pharmacist & independent prescriber
- Built
- Co-founder of SCRUB-in
- Focus
- Evaluate → improve
“The estimated kidney function remains above the usual prescribing threshold, so all medicines can continue unchanged. There is no need for further assessment unless symptoms become more severe.”
Offers false reassurance when dehydration and acute kidney injury are possible.
Assess the clinical picture and the trend in renal function before endorsing continuation.
Clinical judgement for the AI development cycle.
From early prompts to product quality assurance, I help teams find clinical failure modes, improve how their systems respond and evaluate what changed. The output is practical work your product team can use.
Clinical evaluation & red-teaming
Challenge responses with realistic medicines and prescribing scenarios. Identify unsafe reassurance, omissions, unsupported claims and missed escalation, then document the clinical risks.
01 / EvaluatePrompt & workflow optimisation
Examine instructions, conversation flows and safeguards. Propose or implement scoped improvements, then retest against the original failure and new edge cases.
02 / ImproveClinical test sets & scoring rubrics
Build realistic test cases, assessment criteria, error taxonomies and repeatable checks. Measure meaningful clinical behaviour, not just whether an answer sounds convincing.
03 / MeasureMedicines-safety expertise
Apply clinical pharmacy and prescribing judgement to medication information, patient counselling and decision-support content where errors can have real consequences.
04 / SafetyOngoing clinical product input
Work alongside product and engineering teams on defined projects, review cycles or advisory work as a clinical counterpart who understands AI systems.
05 / AdviseRisk registers, revised prompts, case libraries, scoring rubrics and before-and-after evaluation notes. Scope and outputs are agreed for each project.
Discuss a projectI've worked on the problems your team faces.
I co-founded and helped develop SCRUB-in, an AI-powered clinical simulation and OSCE practice platform. That means my experience goes beyond reviewing an AI answer: I've worked on system behaviour, examiner prompts, scoring logic and difficult edge cases.
- 01
Patient and examiner prompts
Structuring conversational behaviour, examiner instructions and case-specific responses.
- 02
Clinical rubrics and safety logic
Defining assessment domains, scoring criteria and critical safety-failure handling.
- 03
Investigation and error correction
Analysing transcripts, identifying incorrect failures and fixing unintended behaviour.
- 04
Iterative regression testing
Rerunning scenarios after prompt and logic changes to check consistency and unintended consequences.
See what the work looks like.
An evaluation should give your team something to act on: the finding, its clinical rationale, a targeted recommendation and the next test.
Explore a complete demonstration diagnostic with a defined brief, three fictional responses, prioritised findings, source links and a proposed safeguard.
Read the demonstration reportSynthetic material, not commissioned work or evidence of measured model improvement.
- Scope and supplied materialWhat is being assessed, for whom and within which limits.
- Prioritised findingsTrace each concern to an output and its clinical rationale.
- A targeted recommendationOne proposed instruction or safeguard for the product team.
- Next testing prioritiesWhat still needs checking before making a performance claim.
Clinical credentials. Product-side experience.
I'm Rob Rumney, a UK-registered pharmacist and independent prescriber. I hold a first-class Master of Pharmacy (MPharm) degree and a postgraduate diploma in clinical pharmacy. My clinical background includes hospital pharmacy and previous specialist oncology practice, with day-to-day experience of medicines safety, clinical decision-making and prescribing.
As a co-founder and hands-on developer of SCRUB-in, I've worked on prompts, clinical cases, scoring rubrics and testing. The product experience above shows how that translates into practical evaluation and improvement work.
You work directly with me. We agree the clinical scope, access, confidentiality, fee and deliverables before starting. Specialist questions outside my competence are identified during scoping.
- Degree
- First-class MPharmMaster of Pharmacy
- Postgraduate
- PGDip Clinical PharmacyAdvanced clinical pharmacy study
- Registration
- UK-registered pharmacistIndependent prescriber · registered as Robert David RumneySearch the GPhC register (opens in a new tab)
- Clinical work
- Hospital pharmacyIncluding previous specialist oncology practice
- AI platform
- Co-founder, SCRUB-inClinical simulation, prompt design, scoring and testing
SCRUB-in is cited as evidence of product-development experience, not as a client or endorsement. Rumney Clinical is independent of my current and former employers.
A practical way to work together.
Not every team needs the same thing. We define the clinical problem, the required access, the deliverables and the limits before starting.
Define the question.
Tell me what the AI does, where it is failing and who uses it. We agree clinical scope, confidentiality, fee and deliverables.
Investigate & refine.
Review scenarios, model behaviour and relevant instructions. Where commissioned, propose or make targeted improvements.
Retest & document.
Where included, run agreed cases again. Hand over evidence of what changed, remaining uncertainties and priorities.
Start with one clinical AI problem.
The Clinical AI Diagnostic brings clinical review and practical improvement thinking together in one small, fixed-scope engagement. I'll examine a defined medicines or prescribing use case, identify clinically important weaknesses and recommend a targeted next step for your development team.
Need something broader? End-to-end evaluation, prompt implementation, larger test sets and before-and-after retesting are scoped and quoted separately.
Explore project scopes and feesA focused review or a longer project.
Start with the diagnostic or commission a larger engagement directly. These indicative budgets help you plan; the final scope and fee are agreed in writing.
Focused evaluation
One clinical workflow, a small agreed test set and prioritised findings with a handover.
£750–£1,500Evaluation & improvement
Baseline assessment, a limited agreed change and a documented comparison on repeat and new cases.
£1,500–£3,000Ongoing clinical input
A defined monthly allocation for review, test design and product discussions.
£600–£1,200 / monthFees vary with clinical complexity, evidence volume and access required. Any VAT, timing and revision limits are confirmed before work starts.
Compare engagement scopesThe practical details.
Have a different clinical AI project? Send a brief and I'll assess the fit.
What clinical AI work do you take on?
Medicines and prescribing-related AI evaluation, prompt improvement, clinical test-case design, scoring rubrics and review of patient-facing clinical content. I check each request is within my scope of competence before accepting it.
Is this a regulatory or DCB0129 sign-off?
No. These are scoped clinical evaluation and improvement services, not regulatory approval, a medical device assessment or formal clinical safety certification. Your organisation remains responsible for its own clinical governance.
Can we send patient information?
Please don't send identifiable patient information through the enquiry form or for the Clinical AI Diagnostic. Synthetic or suitably anonymised scenarios are preferred. Sensitive-data work needs separate agreements and safeguards.
How fast is delivery?
Timing is confirmed in writing once I've seen the material, scope and my availability. I don't promise an automatic turnaround before reading the brief.
Can you work on prompts, not just review answers?
Yes. A separately scoped project can include prompt and instruction review, proposed revisions, clinical evaluation scenarios and before-and-after testing. The £195 Clinical AI Diagnostic includes one proposed targeted prompt revision or workflow safeguard, but no live implementation or validation retesting; larger projects can include both.
What have you built with SCRUB-in?
As a co-founder and hands-on developer, I have worked on AI patient-simulation and examiner prompts, clinical cases, scoring rubrics, detecting incorrect failures, reviewing transcripts and iterating on model behaviour. This is practical product-development experience, not a claim of formal certification.
Can you take on a larger project?
Yes, subject to scope and capacity. Tell me about the model, clinical use case, access needed and intended deliverables, and I can discuss an appropriately priced proposal.
What does the £195 Clinical AI Diagnostic include?
One agreed medicines or prescribing-related clinical AI use case, one supplied prompt or workflow, and up to three representative AI responses of 500 words each. You'll receive a prioritised clinical risk and failure-mode assessment, one targeted prompt or safeguard recommendation, and a concise written diagnostic with evidence and next testing priorities. This is not a complete system audit or a validated performance improvement.
How does payment work?
Fee, invoicing and payment terms are agreed in writing before work starts. This website doesn't take payments.
Building clinical AI? Let’s talk.
Tell me about the product, the clinical risks that concern you and what support your team needs. Evaluation, prompt refinement, test design and ongoing input can all be scoped around the real use case.
Before you enquirePlease don't include patient identifiers, confidential datasets or commercially sensitive details in a first message.
Prefer to write directly? robertrumney1@gmail.com, or connect on LinkedIn (opens in a new tab).