Clinical AI, built from the inside out.
I am a doctor who builds clinical AI. I work inside teams designing and deploying it, where clinical judgement, safety, and product thinking have to sit in the same head as the build.
Clinical expertise, embedded in the build.
I take on consulting and fractional engagements with teams building clinical AI. I sit inside product and engineering, not outside sending advice in. The work spans four areas that usually need four different people.
Building clinical AI tools and workflows
Designing and building the clinical logic an AI runs: conversational workflows, decision points, escalation rules, variable capture, and evaluation criteria. Built and iterated in the platform, not described in a document.
Clinical safety
Clinical Safety Officer work under DCB0129 and DCB0160. Hazard identification, controls, evidence, and residual risk. Safety designed into the system from the start, not bolted on before go-live.
Clinical product thinking
Translating clinical need into something a team can build, and judging when a feature is safe enough to ship. Keeping up in a field where models, regulation, and use cases all move every quarter.
Validation and evaluation
Structured ways to test whether a clinical AI system does what it should and fails safely when it does not. Evaluation criteria, audit methodology, and the documentation that stands up to scrutiny.
Things I have built to show the approach.
These are demonstrations of how I think and what I can build, not commercial products. They are prototypes and worked examples, put together to make the approach concrete. If something here is close to what you need, the real version gets built with you.
Hazard catalogue
A structured library of reusable clinical safety hazards and controls, organised by what a system does. A worked example of how I approach a hazard log.
Clinical pathway template
An acute heart failure pathway taken from national guidance down to the level an EHR team would configure. A demonstration of translating clinical evidence into a build.
CSO hazard-log builder
An early concept for turning the hazard catalogue into a tool that helps draft a hazard log. Not built, deployed, or clinically assured. An idea I am developing.
Safe clinical AI needs a clinician close to the build.
Most healthcare AI has been co-pilot: a clinician stays in the loop and catches the mistakes. What is emerging now talks to patients directly, where there is no clinician downstream to correct it. That changes the safety question from a review at the end to a design problem from the start.
Doing that well needs someone who can hold a clinician's sense of what could go wrong with a patient and an understanding of how the system is actually built, at the same time. That is the work I do.
Building or deploying clinical AI?
Tell me what you are working on and where you are stuck. If I can help, I will say how. If I cannot, I will say that too.
Get in Touch