Contents

    Client

    An American healthcare startup studying endocrine diseases prevention and treatment methods. 

    Project Idea

    The Client addressed us to develop an ML-based diabetes management app for CGM (Continuous Glucose Monitoring) devices users.

    The founder of the company has experienced diabetes himself. He wanted to create a multifunctional mobile app that would help diabetics focus on the long term changes to see the full picture of their condition and significantly improve the quality of life.

    Challenges

    Dive into diabetes monitoring metrics without having the disease

    No one in our developer team has diabetes. That’s why the first challenge we met wasn’t related to technologies themselves. 

    Before the project started we conducted deep audience and competitor analysis. We read numerous stories of people living with diabetes to see what issues they meet every day, summarized the benefits of modern diabetes apps, and discussed how to offer to the people more with our new app. 

    Based on the information and the Client’s insights, we decided to develop these graphs for the app:

    • TIR (Time-in-Range): provides an accurate picture of blood glucose over a period of time.
    • Glucose level: shows average glucose level over a period of time.
    • Coefficient of variation for glucose values: a relative measure of variability that indicates the size of a standard deviation in relation to its mean. 

    We created and approved user flow and started developing the main app graphs. 

    Provide high app performance despite its operating large amounts of data

    The second challenge for us was to make the app work without loss of performance. Since the program operates with a very large amount of data, this affects the smoothness of its work. To solve this problem, we used a database, isolates, and data caching

    Properly organized processes of data acquisition, storage, sampling, processing, and subsequent rendering on the app screen provide smooth transition from screen to screen, when switching between time periods.

    We also learned how to customize the package for drawing charts for our Client’s needs and the app design.

    Test all users’ states and conditions for building accurate graphs

    It was difficult to test all the states that the application can display with only 6 test accounts, one empty account, and only one account with real data in real time. 

    The whole team pitched ideas on how we can implement the task: we discussed each solution with the Team Lead of developers and reproduced by code on simulator situations that could not be achieved with test data.

    Results

    We developed the diabetes management app MVP for iOS and Android in 3 months. The further we moved towards the release, the more pleasant it was to use a product fulfilled with love and care to people. 

    The Flutter framework for frontend helped us to save over 40% of the budget and time the Client planned to spend on development. Due to it, we could use a single codebase for both iOS and Android apps. 

    With Flutter, developers create mobile, web, and desktop apps, using a single codebase for six major platforms and saving a large amount of time and money on the projects. The tech stack helps IT teams release their apps fast and offer high quality to their users at the same time. 

    Our team also formed a large backlog to enrich the app with a unique feature set within 1 year+.

    Solutions

    Recognizable UI and smart UX

    The Client expected us to offer a custom design solution for the diabetes app. Being catchy enough, it should have stayed smart and minimalistic to simplify using the app for people of different ages. 

    We implemented our best medical app development practices to create a unique concept and help the Client stand out among competitors.

    Smooth animation

    With Flutter, we got smooth transitions from screen to screen. The Impeller render engine, accessible for Flutter developers, provides a constant number of frames per second, excellent quality animations, greatly raising the bar for what developers can expect from a multi-platform UI toolkit. 

    Besides, the rendering engine completely eliminates the need for run-time shader compilation, which is a common reason of framerate jitter. 

    Custom feature set

    Building the diabetes management app, we and the Client were aiming to produce diabetics with a tool for monitoring long-term health blood sugar changes. Knowing them, it’s easier to change a lifestyle and a treatment plan.

    Each person with diabetes can easily control their glucose level at the moment or find out TIR per day, but they cannot show the full picture from different sides. The Client decided to create an app that would cover the needs.

    Integrations with popular CGM devices

    Integrated with popular health monitoring apps, the diabetes app we developed gets a user’s CGM data from different CGM devices. Based on the data, the app builds accurate graphs and charts in real time.

    Graphs and charts for controlling diabetes

    We developed five convenient graphs and charts for understanding diabetes. With them, users can analyze their TIR (Time in Range) and its weekly patterns, average glucose level, and coefficient of variation for glucose values. 

    Time intervals to compare your trends

    Users monitor their glucose data for up to 3 months and compare the results to highlight some trends and find out their reasons for each rising and falling. With the data on hand, they plan further lifestyle or medication changes for more effective diabetes treatment. 

    TIR widget on a lock screen

    To make diabetes tracking fast regardless of a user’s situation, we added an opportunity to put the TIR graph on a lock screen. It’s an option that can be used if a person wants. 

    ML-based food recognition

    Our ML experts trained a neural network that can later be applied to photos of food. With the functionality, a person can learn more about favorite dishes and their nutrition value to form a healthy ration.

    To create the neural network for food recognition, we chose Catalyst, a PyTorch-based library widely used for ML research writing. Then converted the neural network to the TensorFlow Light mobile format to connect the model to the iOS and Android apps.

    Smart push notifications

    If any reading is critical, and a user’s health is in danger, the app sends push notifications to warn about it. Thanks to it, a user can take measures immediately, avoid hypoglycemic coma, and other life-threatening conditions.  

    Push notifications are based on target TIR and glucose level readings

    Final words

    Today, the app can be downloaded from the App Store and Google Play. Over 5 thousand users tried it for the first months after release and positively described their experience.

    We’ve been inspired to be chosen as an IT team for this significant project, and we greatly appreciate the opportunity to work for one of our favorite niches. 

    Contact us
    Let’s discuss your project together
    CEO photo white

    Vladi Makeew

    CEO of Surf
    Drop a file here or click to upload
      Hidden span