A clinician-led workflow for safer result review, clearer decisions, and reliable follow-up

Clinical teams review a continuous stream of blood tests, imaging reports, discharge summaries and specialist letters - arriving at different times, from different sources, without the context needed to understand their significance quickly.

I designed a clinician-facing workflow assistant that turns this fragmented process into a structured path from incoming result to interpretation, action and tracked follow-up. The product doesn't replace clinical judgement - it reduces noise, surfaces relevant context, and presents practical next steps a clinician can approve, edit, delegate, defer or ignore.

Most results are routine, but each still requires attention and sign-off. Important abnormalities can be buried among low-risk items, and interpretation is complicated by different laboratories, reference ranges, units, incomplete history and missing clinical notes.

Clinicians repeatedly have to answer the same questions: is this normal, abnormal or critical; is the patient improving or deteriorating; why does it matter for this patient now; what's the safest next step; and who is responsible for completing it. The existing workflow put the burden of rebuilding that context, and coordinating the follow-up, almost entirely on the clinician.

I translated an early clinician-authored product brief and evolving stakeholder requests into a coherent product structure and interface - the MVP scope, the end-to-end workflow for every user role, the information architecture, and the confidence and urgency systems that make the interpretation trustworthy rather than opaque.

A major part of the work was turning incomplete clinical logic into a practical UX model, while clearly separating confirmed requirements from assumptions.

The product was designed around a fixed set of principles: clinicians remain the decision-makers; AI output must be explainable and traceable to source information; uncertainty and missing context must stay visible; unsafe comparisons are flagged rather than silently interpreted; high-risk actions require human confirmation; follow-up ownership and timing remain visible; and every action, suggestion and outcome is auditable.

These principles shaped both the interaction design and the information hierarchy throughout the product, rather than being a compliance afterthought.