Turi Saúde
Building Clinical ambient AI grounded in the realities of Brazilian clinics
Challenge: Ambient AI clinical tools were gaining momentum in the U.S., but the reality for Brazilian clinicians, clinics, and care workflows was much different: uneven internet access, varying levels of technology adoption, multiple clinic contexts, heavy administrative burden, and workflows shaped by both private and public health
This distinct reality warranted a product that was specifically designed.
What I did:
-
Conducted interviews with Brazilian clinicians across different levels of tech adoption to understand how clinicians perceived their own value, what workflows energized them, and what felt draining.
-
Translated findings into mapped journeys that included key automation opportunities and value drivers.
-
Drove alignment by crafting visuals (personas, opportunity solution tree, impact matrix). Facilitated prioritization conversations with founders and defined the minimum viable experience, or “skateboard”, with designed flows and screens for engineering implementation.
-
Independently built and deployed a key product feature: full interface, custom AI agent, appropriate guardrails, and a feedback loop to keep it improving. All shipped in Lovable + Claude, currently used by hundreds of clinicians a day.
Impact:
-
Turi’s AI scribe went from idea to live MVP in 6 months
-
Product is live and used by hundreds of clinicians daily
-
The feature I built improved positive product signal from 30% to 66%; 54% of users said the product was useful because it reduced administrative work, and 22% said it freed up more time for patients
Role - Founding Designer, Ongoing Product & Design Consultant
Year - 2025 to Present
Client - Turi.AI
The challenge
Ambient AI clinical tools could not simply be copied from the U.S. market and dropped into Brazilian clinics.
Brazilian clinicians work across different realities: uneven internet access, varying levels of technology adoption, multiple clinic contexts, heavy administrative burden, and workflows shaped by both private and public health systems. So we set out to answer the following question:
Where does automation actually fit into the day-to-day reality of Brazilian clinicians?
Our goal was to understand the full clinical journey and identify where technology would be useful, adopted, and worth paying for.
A High-level view of my research process. Despite the static representation, this was an iterative process where conversations that happened during "mapping" informed deliverables listed under "understanding".
Multi-step, Multi-stakeholder map: No, your internet is not unstable. Unfortunately this artifact is proprietary to Novonate. I chose to share this image here, albeit blurry, to illustrate the complexity of the process I was mapping. As a gauge, each one of the green squares represents a stakeholder within this workflow.
Researching the real clinical context
I started with the experts. I talked with clinicians across the technology spectrum. Some had fully automated workflows using ChatGPT; others carried binders with handwritten notes.
I explored questions like:
-
What parts of the day drain your energy?
-
What parts of the day feel exciting or energizing?
-
What work would you be open to delegating?
-
What feels too high-stakes to automate?
-
How do different clinicians define high-quality care?
-
How do clinicians move between clinics, systems, and patient contexts?
My goal was to understand how clinicians perceive their own value and what activities they would be willing to delegate to automation.
This research process helped uncover that clinicians wanted support in the repetitive coordination and bureaucracy, and grew weary whenever our tool inched too close to clinical decision-making.
Testing different materials and geometries for our high-fidelity prototypes.
I always sought to include the larger group in this design process. This picture was taken during one of our brainstorming sessions.
Organizing the opportunity space and defining "the skateboard"
After the research, we were now faced with multiple plausible paths with different products, buyers, risk profiles, and go-to-market strategies: emergency medicine or private clinics, public or private healthcare, patient experience or clinician workflow, clinical assistance or administrative automation.
In a nutshell, the team needed help deciding:
Which workflow should we automate first, for which clinician, and why?
To help the team make those tradeoffs, I produced a series of visual artifacts that helped the team have strategic trade-off conversations: compare ICPs, rank automation opportunities, and turn a broad AI vision into a focused first experience that the engineering team could act on.
Getting hands-on: document builder FROM RESEARCH TO SHIPPED USING LOVABLE + CLAUDE
At the strategy level, I helped define where Turi should focus. Then I built one of the first must-have workflows myself.
The document builder had been prioritized by the team, but engineering did not yet have time to tackle it. The existing Lovable + n8n prototype was burdensome and hard to adopt, so I rebuilt it in Lovable as a simpler agentic workflow designed around the document-generation task.
I owned the user research, interface, agent behavior, guardrails, and feedback loop. The goal was to ship something useful now, learn from real clinicians, and give the engineering team more confidence for the long-term build.
Since the rebuild:
-
Positive product signal for document fill-out increased from 30% to 66%
-
54% of users said the product was useful because it reduced administrative work
-
22% said it freed up more time for patients
