USER RESEARCH
OVERVIEW
Making AI Feedback Feel Personal
ROLE
Product Designer
TEAM
2 Project Leads
2 Designers
4 Developers
TIMELINE
Sept 2025 - Dec 2025
TOOLS
Figma
FigJam
STATUS
Handed-Off
Contributions: I focused on redesigning the Learning page of the Performance Review Dashboard and designing an AI-powered learning assistant that helps students improve after practice clinical cases.
CONTEXT
Understanding how to talk to your patients in order to give them the correct care is hard, so you need to practice it.
MedSimAI is a research-backed platform used by 400+ medical students across 6 partner institutions including Weill Cornell, Yale, Ohio State, and University of Virginia, to enhance clinical communication skills with AI-simulated patients. This project is backed by IRB-approved research and we worked directly with real medical students on an already in use active platform.
The problem space our client, Yann Hicke, brought to us was “Redesign the Performance Review Dashboard”. The Performance Review Dashboard (PRD) was what students saw after completing a practice clinical case and where they went to improve on their weaknesses. This dashboard had seven extensive sections, each substantial enough to be its own project. So before we committed to a direction, we needed to conduct research to tell us where the real pain points were.
PROBLEM SPACE
Redesign the Performance Review Dashboard
*But lets narrow this down first
USER RESEARCH
In-person learning vs. AI assisted learning
In order to gain a better understanding of user pain points my co-designer, Aiden, and I conducted a total of 5 user interviews together with medical students of all levels. However, after conducting 2 interviews we were no longer able to participate in the interviews due to IRB policy, so we gathered notes from interview recordings and transcripts moving forward.
After analyzing all interview notes from interviews we found 3 main insights:
Interpersonal Feedback
Users are more likely to improve a specific skill when they received feedback about it in person
“I felt when they were giving me feedback that they had context about who I was.”
Expand Medical Knowledge
Users want to improve on diagnosing by expanding their medical knowledge
"I did leave wondering if I'd really gotten to the right differential... it could be a little bit more robust... for the content of the diagnosis from a medical standpoint."
Maximize Time
Users value in depth feedback but also efficiency and making the most of their time on the platform
“I think we're just pressed. I personally feel pressed for time sometimes.”
To back up what we found in interviews, I compared MedSimAI to Quizlet and tried the platform myself by completing a practice case. Through this I found that Quizlet provides a greater range of study tools for their different types of users. After trying out the platform myself I found that students had to rely on written feedback to grow which was hard to scan quickly and some interactions required extra clicks for little payoff.
PROBLEM STATEMENT
How might we give medical students feedback that feels personal and helps them build medical knowledge efficiently, without needing an instructor in the room?
IDEATION
Choosing the most impactful and feasible features and bringing them to life
We came up with a ton of ideas but had to choose impactful MVP features that addressed all insights. Breaking off into our own features, I focused on expanding MedSimAI’s learning methods such as quizzes, inspired by the gaps I saw in my Quizlet comparison.
However, when I proposed it to our client, he pushed back...
AI chatbot
Can adapt to a students specific stage of training rather than being a one side fits all tool
Tailor feedback for each student and allow them to challenge the AI’s feedback directly
Stay focused on clinical skills
With a new direction we then explored our ideas visually. While exploring, the question of whether the chatbot should be accessible in every page versus in its own page came up. I decided on placing it on every page as users could have questions specific to each page at anytime and being able to reference the contents of that page easily.
CENTRALIZED LEARNING HUB
Making a centralized learning hub for medical students
Our client had vocalized case recommendations as a direction he wanted MedSimAI to grow toward. Helping students find cases tied to the specific clinical skills they are weakest in, instead of choosing random cases. I designed toward that vision within the Learning Hub where students can see their own reflections after a practice case, areas to improve on, case recommendations for the skills they want to practice, and access to the AI chatbot, all in one place.
DESIGNING AN AI CHATBOT
Making an adaptable AI chatbot feel personal and more human
Below is my first iteration of the chatbot I presented to our client, it was designed before I understood what it meant to make a chatbot feel personal rather than generic.
After meeting with our client and receiving feedback from fellow designers, I realized I needed to...
Dive deeper into the types of conversations MedSimAI users would be having
I needed to learn about AI tools more in depth before doing more iterations
Client Feedback
Our client gave me a few points to think about for this AI tool:
What do different conversations look like?
Should the chatbot be a standalone page or a side by side chatbot?
Whats the user experience I want to create for the medical students based on different conversations?
How much information in the thread is shown during the chatbot and how can users easily find past information when the conversation gets long?
I took a step back and after reviewing user interview notes, I decided on 3 main modes users can “enter” when they begin a chat.
Medical Tutor
Supportive tone
In-depth explanations
Research
Direct information without extra jargon
?
Quick Support
On-platform support
Maximize time
After grounding myself on these types of modes and user and client needs, I began iterating on designs for the chatbot once again.
Displaying modes intuitively with suggested prompts
Students can easily ask questions with the familiar interface and prompts tailored to each mode.
What represents a Medical AI Chatbot within an AI platform?
Originally, the name of the chatbot was “Medbot”, but thinking about our insights I realized it was important to make the interaction with the chatbot more humane. I iterated on the logo using MedSimAI’s current logo so that it does not feel so robotic like the original icon. I came up with the name “Meddie”, however I realized it was too informal for a professional platform such as MedSimAI, leading me to the name “Lumen”
Original
Re-design
Rather than generic AI feedback, the chatbot has full context of a student’s specific clinical case (transcript, scores, case details), closer to how a professor or attending physician who knows that medical student, their strengths and weaknesses. Naming the chatbot “Lumen” instead of “Medbot” as well as displaying a visual reminder that “AI makes mistakes” was to make the chatbot feel more personal and trustworthy rather than robotic and forced, addressing our Interpersonal feedback insight.
THE SOLUTION
A way to learn like you had a professor right next to you to help
Catered modes for different types of students
Medical Tutor Mode
A personal tutor helping and supporting students at their own pace.
Research Mode
Quick and direct information on medical information and clinical steps.
Quick Support Mode
Directions and direct responses to in-platform questions.
Making sure the learning experience wasn’t interrupted
Full screen chatbot for an immersive experience → no distractions on the screen while learning
History option to access past conversation topics → no scrolling endlessly to find something.
REFLECTION
What I knew then vs. now
Sometimes, if it works, you don’t have to change it
Starting out I wanted to design a chatbot that was different from other chatbots, but I realized that there was a reason so many followed similar practices. So even wanting to be different is good, sometimes you don’t have to change something especially if it works.
Polishing visuals are the last things to refine, not the first
We had to deliver hi fi designs from lo fi in just 1 week. At first, I spent too much time on refining small details on screens whose flow I hadn’t fully decided on yet. This led to needing to work even faster and spend more time in order to finish the designs. So always refine the flow and major screens then spend time on polishing visuals.
This was the first time I had worked with a real client, I didn’t know how to pitch my designs or defend them. Months into the project, I had learned how to do both and more importantly learned how to think about user needs and client needs at the same time, rather than choosing one over the other. Having to defend my decisions out loud, in front of someone who could directly reject them, made me truly think about whether my designs were good and solved the problem.
It was a pleasure to be a designer on MedSimAI and grow so much in just 3 months.
In our final meeting before handoff, Yann approved Lumen and the chatbot experience and we handed off our designs to the engineering team of MedSimAI. I’m excited to see how Lumen shapes the way students build clinical confidence once it is built!
Want to learn more? Feel free to reach out, I’d love to chat with you more deeply about my design process!