Private case study
This project is shared selectively. Enter the password below to view the full story.
Want to connect directly? Book a meeting or browse selected work.
Product Strategy · Design systems · AI UX · Human first
A speculative design brief. What I’d build if I joined.
This page is grounded in people first, friction second: the joy clinicians find in caring for people, and the erosion they feel from admin and cognitive load. My instinct is to strip pain so doctors can do what they trained for—by mapping where friction lives, asking why it exists, then designing it out with clarity and humility.
Built from Heidi Health’s public positioning and metrics only—no insider knowledge. A portfolio exercise in how I’d frame product design leadership for a clinical AI care partner.
In 18 months, Heidi returned over 18 million hours to frontline clinicians. 73 million consults. 200+ specialties. 116 countries. That’s not traction — that’s a clinician-led movement.
The product-market fit was found through a freemium AI scribe that just works. Now the mission is expanding: from documentation, into Tasks, Evidence, Comms, Ask Heidi, and a brand new 21g wearable — Heidi Remote. From tool to care partner.
That expansion is where the design challenge gets genuinely hard. Because each new surface, each new capability, each new workflow is another moment where the product either deepens clinical trust — or quietly erodes it.
$65M Series B
Oct 2025 · $465M valuation
18M+
Hours returned to clinicians in 18 months
2M+
Consults supported every week
110+
Languages across 116 countries
Clinical AI only matters if it fits the rhythm of real care: interruptions, uncertainty, empathy under pressure. I’d keep research and design rituals close to those rhythms—shadowing, co-design, and blunt feedback from people who don’t owe us optimism.
Tom’s bar for clinical software is unforgiving: if it adds cognitive load, it fails. These principles keep the work honest when the roadmap gets loud.
The best clinical tools disappear into the flow—documentation, suggestions, and safety nets feel like part of the conversation, not a billboard on top of it.
Trust in clinical AI is earned in the room, not declared on a slide. I’d design so clinicians can always see what Heidi is confident about, where it’s less certain, and what each output is grounded in—provenance and humility made visible, not buried. Transparency belongs in the flow of care, not in the fine print.
A disciplined system is a safety strategy: predictable patterns, audited components, and a single language for risk-critical states across surfaces and specialties.
How I’d organise design work at Heidi: trust in the UI, truth in the field, and a system that keeps both honest at scale.
Sequencing matters: earn context before you earn roadmap. Three horizons, one thread—reduce cognitive load for clinicians and teams shipping alongside them.
Weeks 1–2
Map reality; align on what “trust” means in the product
Weeks 3–6
Tighten the design system around clinical risk states
Months 2–3
Ship learning loops: measure trust where it’s earned
Most SaaS design systems are built for speed and consistency. Heidi's needs something more: it needs to encode clinical confidence.
When a clinician glances at an AI-generated note, they're not just reading text — they're making a rapid trust assessment. Is this right? Is this mine? Would I sign this? The typography, density, hierarchy, and micro-interactions in that moment either support that assessment or create friction.
Visual language that communicates AI certainty levels without breaking clinical flow — not warnings, not alerts. Quiet, precise, contextual.
Every AI output must feel immediately, obviously editable. The clinician is always the author. The system should make that felt, not just stated.
Clinical environments are cognitively loaded. The system should give information the space to be read, not optimise for feature density. Less on screen. More in mind.
From web to mobile to Heidi Remote — the experience should feel continuous. A clinician moving from desktop to ward round shouldn't relearn anything.
Post-consultation handoff pattern
This pattern comes from my NSW Aged Care work with Microsoft, where the handoff after each consult was the highest-cognitive-load moment. The clinician is transitioning out of presence with a patient and into action mode. What the system surfaces in that 30-second window shapes whether AI feels like a partner or just another task list.
Pattern observed in prior aged care project
Generated note + tasks listed sequentially. Cognitive load transferred, not reduced.
Explorations
High craft standards and psychological safety aren't opposites. The best design teams I've led held both at once — and the quality of the work was the proof.
I'd set the bar through my own work first. Show up in the Figma file. Give honest critique. Be wrong sometimes in public. Make it normal to say this isn't good enough yet — including about your own work.
The thing I've learned about designers in clinical contexts: they care more than in almost any other domain. They know what's at stake. The job is to give them the conditions to act on that care — clear principles, fast feedback cycles, and the autonomy to own their work properly.
Heidi is operating at the intersection of care, software, and societal trust. If I were leading product design, my job would be to keep the work grounded in real clinicians, real time pressure, and real accountability—while building a system that scales without turning cold.
Portfolio contact: shaziakazi.com
If you’re hiring for healthtech product design—or want to pressure-test a brief like this—I'd welcome the conversation.
Focused on practical outcomes
Based in Sydney, working globally
