Managing Type 2 diabetes can be tricky because patients need to take the right amount of insulin at the right time. If they take too much or too little, it can lead to serious health problems. Right now, many people rely on manual calculations or guesswork, which is not always accurate.
Our project aims to make insulin delivery smarter and easier by using a Raspberry Pi (a small computer) and machine learning. Instead of relying on real-time sensors, we use historical data (past glucose and insulin records) to predict how much insulin a person needs. The Raspberry Pi will then control a small pump to automatically deliver the right amount of insulin (using water in our prototype).
Many individuals with Type 2 diabetes face challenges in determining the correct insulin dosage, as it depends on factors like food intake, physical activity, and overall health. Delays in administering insulin can result in dangerous fluctuations in blood sugar levels, leading to sudden spikes or crashes. Additionally, the risk of miscalculating the dosage, whether too much or too little, can cause serious health complications, making effective diabetes management both difficult and crucial.
- Syd Ainsworth - (p01: Methods, Findings, Conclusions, Caveats) + (p02: Wireframes in Figma, Methods, Conclusions, Caveats) + (p03: Initial prototype construction and frame wiring, IRB submission, Study spreadsheet formatting, Methods editing, Findings, Caveats)
- Sri Ramani Thungapati - (p01: Methods, Findings, Conclusions, Caveats) + (p02: Findings, Conclusions) + (p03 - Introduction, Methods, Protocol Script, Refining Wireframes)
- Elizabeth E Rodriguez - (p01: Executive summary, Introduction, and Design Artifacts) + (p02: Wireframes, Executive summary, and Introduction) + (p03: Prototype Design, Executive summary, and Conclusions)
Executive Summary
- Understanding User Needs – We identified that diabetes management goes beyond insulin delivery. Users need support in tracking food intake, mood, and overall well-being, making the system more than just a device.
- More Than Just a Device – Our system includes features like ChatGPT integration, allowing users to ask health-related questions such as “How much sugar is in a Pink Lady apple?”, helping them make informed dietary decisions.
- Mood Diary for Tracking Well-Being – A built-in mood diary allows users to log their feelings and health status, giving them a better understanding of how their mental and physical states impact glucose levels.
- Competitor Analysis – We researched existing insulin pumps and diabetes management apps, identifying gaps in automation, personalization, and user engagement that our system aims to improve.
- Smarter Insulin Predictions – Unlike traditional pumps, our machine learning model analyzes past data to predict insulin needs more accurately, reducing the risk of overdosing or underdosing.
- Less Work for Patients – By automating insulin delivery, we eliminate the need for manual calculations and adjustments, making diabetes management simpler and more stress-free.
- Prototype is in Progress! – We have purchased equipment and are actively building the system to control the insulin pump (simulated with water) using the Raspberry Pi.
- Personalized for Each Person – Our system is designed to learn from individual glucose patterns, making insulin delivery customized for each user rather than a one-size-fits-all approach.
- A Step Toward Better Diabetes Care – Through AI and automation, this project has the potential to enhance diabetes management, improving patient confidence, convenience, and overall health outcomes.
Executive Summary
- Onboarding: Sign In and Sign Up screens were clearly labeled and intuitive.
- Progress Feedback: Users received visual confirmation (screen transitions, success messages) when performing actions.
- Dashboard Layout: Users found glucose levels were easy to locate and understand.
- Alert System: Notification flow was recognizable and encouraged user response.
- Navigation: Setup steps used straightforward buttons and transitions.
- Accessibility Concerns:
- A user found that the app’s navigation was difficult for our visually impaired persona, James, as it did not fully support speech-to-text features. However, this issue was due to user error, as most devices offer these apps.
- Bolus Rate Confusion:
- Our persona, Tyler (caregiver) was unsure how to adjust bolus settings for his daughter.
- Setup Completion Ambiguity:
- Final setup screen lacked direction or confirmation on how to proceed to the dashboard.
- Information Gaps:
- Our persona, Tyler had difficulty adjusting insulin settings because of medical jargon and a lack of helpful explanations in the app.
- Introduced tooltips and help icons for bolus settings to assist caregivers.
- Updated final setup screen with a “Continue to Dashboard” button and success message.
- Clarified error messaging during login to help users identify incorrect credentials.
- Ensured alert notifications include confirmation and clear call-to-action (e.g., “Acknowledge,” “View Details”).
Executive Summary
In this sprint, we tested our SmartInsulin mobile app prototype with real people. We wanted to see how easy it is for users to get started with the app, use its main features, and share their thoughts on what worked and what didn’t.
The goal was simple: make sure our app actually helps people managing Type 2 diabetes (either for themselves or someone they care for). We watched how participants used the app in common, real-life situations and asked them to talk through what they were thinking.
We asked users to try six tasks using the app:
- Set up the app and pair their insulin pump
- Export blood sugar reports as a PDF
- Log an unusual blood sugar reading
- Ask a dietary question using the AI chatbot
- View glucose and insulin trends
- Monitor someone else’s diabetes data (as a caregiver/parent)
After each task, they rated how easy it was (1 = Very Hard, 5 = Very Easy) and shared their thoughts.
- Most people completed tasks easily — especially if they were familiar with health apps.
- Trend graphs were helpful and clear for many users.
- Note-logging and chat assistant features were seen as useful and practical.
- Caregiver monitoring was a big hit for those supporting loved ones.
- Setup needs better guidance for first-time users.
- Exporting reports was hard to find — users want clearer labels and confirmation.
- Note logging could be faster — fewer taps and more pre-filled options.
- Chat answers were good, but users wanted sources and safety disclaimers.
- Some users were confused by trend data — we need better explanations or tooltips.
Thanks to everyone who tested the app and helped us move one step closer to a truly helpful tool for diabetes management!