Fellowship Bio

Clinician with 10+ years’ NHS experience, building a career at the intersection of clinical practice and AI. MBBS (KCL), MPhil in Medical Epidemiology and Statistics, Clinical Safety Officer certified. Completed the NHS Fellowship in Clinical AI, building an LLM pipeline to extract pathology data.

Pathology reports are full of vital information, but almost all of it sits as free text. When a clinician needs specific information from a melanoma diagnosis, they have to find and read the report themselves, and that time cost adds up fast across a pathologist’s workload. My project asked if a large language model could read a melanoma pathology report the way a pathologist does, and extract the 20 essential fields defined by the Royal College of Pathologists’ melanoma dataset into clean structured output, freeing up time for diagnosis and patient care. I also wanted to test whether this could be built the way most NHS Trusts realistically would need to: locally, on a small model, without fine-tuning or cloud APIs. As co-lead, I designed, built and evaluated the tool in Cellular Pathology at University Hospitals Southampton. I use LLM-assisted coding to build the pipeline from scratch: a rule-based screen, an LLM pass to separate specimens, and a second pass extracting the fields, with post-processing to validate outputs. It runs on a local GPU, so no patient data leaves the Trust. I designed the evaluation framework, measuring field-level accuracy, F1, PPV, NPV and Cohen’s kappa, and stratified fields into confidence tiers. I co-authored the DPIA and am writing a methods paper for peer-reviewed submission. Development moved from 51.4% accuracy at baseline to 75.5% after refinement, validated on held-out cases each stage. The latest evaluation reached 84.5% mean field-level accuracy and 89.7% PPV, with three fields now reliable for minimal oversight. Next steps: improving the remaining fields, growing the dataset, a multi-model approach, and a user-facing interface with the governance that clinical use needs.

Fellowship Testimonial

This fellowship allowed me to learn the fundamentals of AI in healthcare and apply them to a live project. As co-lead of a brand new project, I had freedom to shape its direction, but also responsibility for solving every problem, starting with no budget, limited infrastructure, and no precedent in our department. Over the year I built relationships with clinical and non-clinical colleagues, and our hospital’s AI Oversight Committee, gaining practice pitching project goals and finding people willing to support the work. These connections led to a partnership between Cellular Pathology and the Southampton Emerging Treatments and Technologies (SETT) Unit, with supervision from the Trust’s Clinical AI Lead. Through SETT I learned to navigate the Trust’s AI governance pathway, providing documented evidence of compliance and patient safety to our AI committee. I gained confidence building something from scratch in healthcare, communicating with non-clinical IT colleagues outside my usual work, and translating research skills into building real tools, including the basics of software engineering. The technical milestones were satisfying, but the biggest reward was the people. The fellowship’s culture of open collaboration was energising, and being surrounded by others tackling the same problems gave me real optimism. I plan to pivot fully into health tech, likely medical consulting or product/workflow transformation. I look forward to where AI in healthcare is going and the opportunities ahead.