I take a clinical problem from the first conversation to a working interface. I bring the questions of a pharmacist, the rigor of an engineer, and the willingness to build.
From prescribing complexity to a working calculator.
Inspect a source without losing your place.
Alkalinizing medications come in different formulations and units. Clinicians need to compare their alkalinizing, sodium, and potassium contributions without losing track of what each medication adds.
The interface needed to make those relationships readable while leaving the clinical decision with the physician.
I proposed the roadmap and Figma mockup for physician review, structured the medication information and calculation logic, and built the React/JavaScript interface.
I worked with a nephrologist and a pharmacist, incorporated feedback from roughly ten nephrologists, and added guidance and downloadable summaries. I stayed involved through testing and documentation.
Discuss the clinical problem and establish the calculation approach in a spreadsheet.
Propose a roadmap and a Figma mockup for physician review.
Implement the interface and check data, calculations, and displayed outputs.
Use specialist feedback to clarify the interface and guidance.
Medication search fits the task of selecting a known product. It reduces the need to scan a long list.
Tradeoff: a clear selection flow still needs readable formulation and unit information. Finding the name is not enough to prevent choosing the wrong variant.
Individual medication cards stay distinct from aggregate contributions, so a clinician can inspect where a total comes from.
Tradeoff: more visible information can add clutter. Grouping and hierarchy matter more than removing useful context.
An integrated guide and downloadable summary support understanding and discussion beyond the calculation itself.
Tradeoff: an export is a snapshot. It should not imply that information is still current or replace the clinician’s review.
I would investigate clearer formulation labels, a final review step, and explicit data-version information. I would observe clinicians completing the same tasks before choosing what to change. These are proposed next steps, not features or outcomes already delivered.
Source: my PharmD thesis, pp.33-34 (project), 45-47 (prototype), 63 (implementation), 64-68 (testing), 69-73 (feedback and improvements). Screens reproduced from the thesis.
A source should add clarity without stealing the reader’s place.
How might a clinician inspect the support behind an answer, compare its scope, and return to the same passage without rebuilding their mental context?
This independent concept was developed for this application with AI assistance. Its clinical framing draws on my pharmacy training and CALALCANI experience. The scenario is illustrative; clinician testing is a proposed next step.
Population, setting, publication date, and what the source actually measured.
A guideline may address a broad population while a study examines a narrower group. Check whether the source applies to the question before treating its conclusion as transferable.
A finding about one endpoint does not establish every outcome that matters to a clinician or patient. Keep the measured outcome visible when comparing sources.
This demo helps inspect evidence. It does not resolve a clinical disagreement or recommend treatment.
Select A or B to inspect a source.
On desktop, sources open beside the answer. The cited passage stays highlighted. On mobile, the source becomes a focused dialog with a clear return action.
Tradeoff: a split view reduces answer width. The layout changes when it can no longer keep both columns readable.
The source panel separates excerpt, population, and limitations. It avoids an unsupported confidence percentage or a badge that implies verification.
Tradeoff: this adds reading. Progressive disclosure keeps the essential context visible and lets readers expand the rest.
A missing source stays visibly unavailable. The answer remains accessible, and the reader can retry or return without silently substituting another source.
Tradeoff: the failed path feels less seamless. That friction is preferable to suggesting evidence was checked when it was not.
I would begin with six clinicians as an exploratory sample, alternating task order between a separate-source view and this contextual view.
Measure task completion, lost-context events, misunderstandings, and time to return. Six participants would reveal usability issues, not establish clinical safety or generalizable outcomes.
The working prototype uses HTML, CSS, and JavaScript. The design specifies source states, reading hierarchy, responsive behavior, and keyboard interaction.
I'm a doctor of pharmacy and biomedical engineer based in Lyon. At Onicor, I analyzed usability data in Python, examined methodological limitations, interviewed two podiatrists about adoption barriers, and created scientific communication materials.
My pharmacy training included laboratory practicals. It taught me to question the measurement, the method, and the conclusion. I bring that habit to product work: check what was measured before deciding what the result means.
I work in French and English. I'm looking for a demanding team where I can learn quickly, get direct feedback, and take responsibility for a useful product.
