GlucoSense:
AI in Diabetes Care
A simple way to map out your life and make sense of your glucose data

Project type
Sponsored project
(Microsoft x HCDE)
Duration
5 months (Jan - May 2024)
Team
Helen Li
Jeeah Eom
Michelle Wu
Tiffany Li
Role
Research
Design
Prototyping
Usability testing
Logo design
Animation creation
Tools
Adobe Suite
Figma
ProtoPie
OVERVIEW
Background
The global population diagnosed with diabetes is increasing at a significant rate. In the US alone, 38.4 million people have diabetes, which is 11.6% of the population. Simply put, one in every ten individuals is living with diabetes. Therefore, this project aimed to make a meaningful impact in this field.
Problem
Successful diabetes management relies on interpreting blood glucose patterns, which requires contextual information such as physical activity, diet, and sleep. However, it can be burdensome for patients to consistently record this information, making it difficult for providers to interpret their data and recommend treatment changes.
Solution
Our solution to this problem is GlucoSense, an AI-powered tool that offers seamless logging of contextual information, maps it with blood glucose data, delivers activity-based insights, and automatically captures notes from provider appointments for future reference.

Product Video
DESIGN QUESTION
Our project started with a prompt from Microsoft: "How might we elevate patient experience leveraging AI?". We chose diabetes as our focus due to the alarming increase in diabetes diagnoses both in the United States and worldwide.
Through primary research, we realized that interpreting patterns in blood glucose data is key to successful diabetes management. However, one critical aspect of this interpretation — contextual information — is currently lacking. This led us to our design question:
"How might we help patients effortlessly collect contextual data to make sense of their blood glucose levels?"
RESEARCH
We conducted the following research methods to gain a better understanding of our target users and the overall problem space.
Literature Review
Firstly, we conducted a literature review to better understand the diabetes management space, current challenges, and potential opportunities. Through this method, we discovered that diabetes management requires a multi-faceted approach involving medications, diet, exercise, and monitoring. Patients often feel overwhelmed by the abundance of information, making it difficult to take effective action or adjustments.
Competitive Analysis
We also conducted a competitive analysis among 10 competitors to understand the current landscape of AI-powered diabetes tools and identify gaps and opportunities. We found that current diabetes management tools tend to focus on specific aspects like nutrition or exercise, rather than offering a comprehensive approach.

Interview
We conducted user interviews with providers, patients and AI expert respectively.

Provider Interviews
We conducted interviews with 5 endocrinologists to validate the problem space and gain insights into treatment processes and the diverse patient populations being treated. Through these conversations, we discovered that providers often lack the necessary context to understand the reasons behind patients’ blood glucose patterns. This context includes factors such as physical activity, diet, stress levels and sleep. The absence of this information poses challenges for providers when interpreting patient data and recommending treatment changes.
Patient Interviews
We conducted interviews with 5 patients to understand the challenges they face and to learn more about various aspects of their diabetes management. Through these interviews, we found that patients find it burdensome to consistently record contextual information, which is often spread across multiple platforms. Additionally, during consultations, providers review patients’ blood glucose data and suggest treatment plan changes. However, the abundance of topics discussed can make it difficult for patients to remember the details and adjust their routines accordingly.
AI Expert Interview
We conducted an interview with Dr. David Rhew, Global Chief Medical Officer and Vice President of Healthcare at Microsoft, to understand the feasibility, constraints, and potential impact of the AI tool we envisioned. He suggested leveraging AI's strengths in synthesizing and presenting information in an understandable format and emphasized the importance of making data collection from users as easy and frictionless as possible.
DEFINE
User Persona
Based on the research findings, we developed a persona of our target users to help guide our design decisions.

User Journey
We also created a journey map to illustrate how Elvia would use our solution in her day-to-day life.

DESIGN
Next, we planned to design this platform, but before doing so, we established the requirements as guidelines for the subsequent design process.
Design Requirements
01 Seamless user input: make it as easy as possible to collect data from users
02 Personalized insights: offer personalized insights based on patient activity
03 Effortless notetaking: automatically record information from consultations and summarize
key takeaways
04 Transparency & control: offer transparency on how AI generates information and give users
control to edit or correct
Design Process
We took an iterative approach to design by collecting user feedback and improving the design based on feedback received. This led to an overall more refined and user-aligned solution.

Co-design
We visualized our concept using a storyboard and conducted a co-design workshop with 3 participants to get their feedback on our overall concept and involve them in the process of designing a solution that integrates seamlessly into their daily lives.

During the co-design workshop, participants pointed out their concerns and identified problematic areas. Together, we brainstormed solutions to address these issues. The workshop provided several key takeaways, including:
● Offer different ways to input data to minimize user effort
● Offer flexibility in notifications
● Allow input of qualitative info such as events (e.g., traveling, illness), stress levels, and mood
● Give users control (e.g., edit AI-generated content, turn off ambient listening)
Prototyping
Based on insights from the co-design workshop, we began developing a prototype to visualize our concept. We started by creating the information architecture to guide the app design and better illustrate user flows.

According to the information architecture, our app provides 3 main features, as illustrated below:
Feature 1: contextual map + logging options
Patients often forget to record their meals, activities, and other details due to their busy lifestyles, finding it tedious and burdensome. To address this, our solution integrates data from the apps and wearables they already use and employs ambient listening to identify activities based on background noises. This approach helps patients effortlessly map out relevant life activities.
Also, to address gaps that our app might not detect, we provide patients with options to manually input data. They can take photos, record voice notes, or type their entries. For example, patients can use camera feature to take pictures of their meals. Our app will then automatically calculate the carb count and record the nutritional details in their log.
Context mapping


Logging option: take photos

Logging option: type entries

Logging option: record voice notes

Feature 2: personalized insights
While Continuous Glucose Monitors (CGMs) provide some patterns, they are not personalized to each patient. Additionally, patients expressed frustration about not knowing what has changed and what they could do differently based on their current condition. Therefore, we designed our app to offer personalized insights based on patients' logs, helping them better manage their condition.

Feature 3: ambient assist
Patients have regular visits with their healthcare providers, during which they discuss multiple topics. After these consultations, patients often forget key points that were discussed. To address this, our app includes an ambient notetaking feature. This feature allows patients to activate ambient listening during consultations, automatically capturing notes and recording key takeaways for later reference. This helps patients recall important action items or discussion points from their appointments.

User Testing
We then tested our prototype with users through unmoderated testing to gather feedback on the overall concept and identify any areas of confusion. Here are some key takeaways from these 6 tests:
● Some wording was unclear for users (e.g., "ambient assist", "events")
● Offer varying levels of detail in content to align with specific user preferences
● Include an onboarding flow to set expectations with users and provide an overview of the app's
concept and purpose

Iterations
Based on the feedback from user testing, we refined our prototypes and updated our design system to create high-fidelity prototypes and an onboarding flow.
Feature 1: context mapping + logging options


Feature 2: personalized insights


Feature 3: smart notetaker


Onboarding flow
Context Mapping

Personalized Insights

Smart Notetaker

RESPONSIBLE AI
As we design our product with AI, we believe it’s crucial to do so responsibly. Two potential harms it may pose to users are: a) inaccurate information and subsequent user reliance on it, and b) privacy concerns related to data collection and storage. To address these potential harms, we've developed solutions to mitigate them.
To address inaccurate information:
● Implement feedback loop to collect user input on AI-generated information
● Include disclaimers during onboarding and throughout the app UI to caution users
● Allow users the flexibility to modify or delete data
● Highlight potentially inaccurate or high risk information

To address privacy concerns:
● Ensure transparency around data collection and storage processes
● Give users control over what data is collected and stored
