Designing an AI-Powered patient summary tool for oncologists
Context
In oncology, when a new patient enters care, attending physicians must quickly understand their full medical history — often spread across dozens of documents. To support this, our team set out to build an AI-powered tool using large language models (LLMs) to extract and structure key clinical information into a usable summary.
This was part of a broader initiative exploring how foundational models could meaningfully support clinicians in real-world settings.
Skills
UX Leadership / Stakeholder management / Product Management
Interaction design / Visual design / Information architecture / Wireframing / Prototyping / Soft skills / Usability testing / Documentation
Main UX Challenge
Having the technology, team was still getting build and required detailed information was not fully yet available. How do we make an AI product that is useful and usable for attending physicians, without overwhelming them with noise or assumptions, while increasing the UX maturity in the company?
My role
I worked as the lead UX designer, collaborating closely with:
A technical lead (engineer wearing the project lead hat)
An internal medical expert (former surgeon) who co-led our clinical alignment
An external subject matter expert from the oncology field (Head&Neck Surgeon)
I was responsible for driving the design process across a wide range of areas:
Translating ambiguous technical input into early concepts and wireframes
Weekly iteration with our internal and external clinical experts to refine the product direction
Leading and synthesizing 20+ cognitive walkthrough interviews with physicians across departments
Creating mid- and high-fidelity prototypes
Supporting front-end engineers through handoff and implementation
Helping the team align across shifting requirements, unclear data availability, and the absence of formal product leadership
The approach
Cognitive walk-through interviews
To bring clarity to a fragmented ideation phase, I synthesized input from across the team into a tangible prototype that served as our initial product foundation.
I led iterative working sessions - first internally, then with external clinical stakeholders at Netherlands Cancer Institute - using each round to refine the concept.
As a result of this process, we established a clear, minimal-information structure for the patient summary that balanced readability with clinical completeness, setting the groundwork for product implementation.
Example of the wireframe
From wireframes to hi-fi and Design System bases
After conducting over 20 interviews and aligning on the key clinical content, I transitioned from concept validation to designing the actual interface and workflow. The challenge was twofold: presenting dense, text-heavy information in a clear, uncluttered way and doing so without an established brand or product design framework.
I made strategic design decisions grounded in two core principles I developed with a former teammate: clarity over decoration and relevant information and bold design.
I proactively began building a flexible design system alongside the interface, anticipating future needs and edge cases across devices, screen sizes, and browsers.
In parallel, I advocated for the formal adoption of a design system within the company, laying the foundation for scalable, consistent design practices across future AI products.
Project challenges
This was not a case of "AI first, UX later." It was a moving target: product, data, and even goals were still being defined. The biggest challenges came from within the team: technical ambiguity, unclear priorities, and shifting definitions of success.
Design had to act as a sense-making force by helping shape the product itself, not just decorate it.
Impact
The concept I designed became the foundation for the implemented product.
It helped the team align on a shared direction
It shaped the core interface and workflow
It enabled physicians to engage with the idea early and shape it before engineering was fully committed
This work now lives as a core feature in our clinical AI platform — built around a concept that was validated, pressure-tested with real physicians, and shaped with close collaboration between design, engineering, and medical experts.