Diabetes-

Diabetes- turns everyday health logs into meaningful insights, helping people with diabetes understand their bodies, manage medication safely & recognise warning signs before problems get bigger

Comment

Diabetes-

Your doctor sees you for a few minutes every few months. This app watches the weeks in between.

Diabetes- is a bilingual AI-powered diabetes companion for Bangladesh. It helps people understand their condition, monitor how their bodies respond to medication and everyday habits, recognise warning signs, and manage diabetes more confidently.

Inspiration

Taking medication is not the same as understanding your condition.

For millions of people living with diabetes, managing the disease goes far beyond remembering to take a pill or checking blood glucose. It means making dozens of decisions every day: what to eat, how to exercise, when to take medication, how to respond to symptoms, and when something might be wrong.

But there's a dangerous gap between being prescribed treatment and understanding how to manage it.

Some people believe they can compensate for excessive sugar consumption by taking more insulin. Others may take insulin without properly considering their food intake or following their prescribed plan. Many overlook changes in their bodies after starting medication, while others turn to the internet for answers and end up unnecessarily frightened by ordinary symptoms.

We wanted to address that gap.

Because diabetes management shouldn't be a guessing game.

We built Diabetes- to make everyday diabetes management more understandable, more personal, and more informed.

What it does

Diabetes- is a bilingual English-Bangla diabetes companion designed for everyday life in Bangladesh.

It goes beyond basic health logging. It helps users connect their daily habits, medication routines, symptoms, and changing health indicators to develop a better understanding of their condition.

Key Features

  • Medication Response Tracking: Helps users observe changes in indicators such as weight, glucose, and symptoms over time, making potentially important patterns easier to notice and discuss with their doctor.
  • Safer Medication Awareness: Reinforces the relationship between food, glucose, and prescribed insulin routines. It encourages users to follow their individual doctor's instructions rather than making unsafe adjustments themselves.
  • Everyday Health Management: Helps users keep track of meals, glucose, medication, exercise, hydration, energy, and mood.
  • Symptom Intelligence: Users can describe what they are experiencing in their own words. Rather than turning every headache into a frightening diagnosis, the app guides them through structured questions, including whether symptoms are persistent or recurring.
  • Learn: An awareness and myth-busting section that makes diabetes education easier to understand, one chapter at a time.
  • Safety Switch: Provides emergency guidance and a way to alert a care person, alongside relevant emergency helpline information.
  • Doctor-Ready Reports: Organises recorded information into a structured report that users can take to their healthcare provider.

Our philosophy is simple:

We don't turn every symptom into a horror story. We look for patterns, context, and signs that deserve attention.

How we built it

We built Diabetes- using Python and Streamlit, with custom CSS for a responsive, mobile-first interface.

The AI component is powered by Google's Gemma model through the Google GenAI API.

One of our most important architectural decisions was to separate AI assistance from safety-critical logic.

Python handles deterministic rules such as glucose thresholds, emergency red flags, urgency levels, and structured symptom pathways. Gemma assists with interpreting free-text descriptions and mapping them to predefined symptom modules, as well as explaining approved educational information.

We don't let AI play doctor.

The model does not diagnose conditions, determine emergencies, or recommend medication dosage changes. When the AI service is unavailable, the application also has fallback behaviour.

The application supports English and Bangla, including locally familiar foods and language choices, because health technology should be understandable to the people it serves.

Challenges we ran into

The biggest challenge was deciding how to use AI responsibly in a health-related application.

A model that confidently produces an answer is not necessarily a model that produces a safe answer. We needed to prevent the system from inventing diagnoses, exaggerating symptoms, or making medication recommendations.

We addressed this by keeping critical decisions in deterministic Python logic and limiting the AI to clearly defined tasks.

Another challenge was making the experience relevant to Bangladesh. Health education often becomes inaccessible when the language, examples, or assumptions do not reflect the user's everyday environment.

We also had to balance simplicity with functionality: the application needed to be useful without overwhelming someone who may already be dealing with a complicated daily routine.

Accomplishments that we're proud of

  • Built a working bilingual diabetes companion rather than just a conceptual prototype.
  • Integrated Gemma into a practical, constrained AI workflow.
  • Created a structured approach to symptom reporting that avoids unnecessary alarm.
  • Connected everyday health logs with trends that can help users have more informed conversations with their doctors.
  • Designed medication and insulin awareness around individual prescribed plans rather than generic dosage advice.
  • Built a doctor-ready reporting experience and an emergency-support pathway.
  • Made the experience relevant to Bangladeshi users through local language and food context.

Most importantly, we are proud that the project is built around a problem people face every single day, not AI for the sake of AI.

What we learned

We learned that building AI for health is not about making a model sound more intelligent. It is about knowing when it should speak, when it should ask questions, and when it should stay within its limits.

We also learned that the most valuable health insights are not always found in a single reading or symptom. Sometimes, they emerge from patterns across days and weeks.

And perhaps most importantly, we learned that technology cannot replace a doctor, but it can help people become more informed participants in their own healthcare.

What's next for Diabetes-

We want to take Diabetes- beyond a companion for logging and awareness and develop it into a more comprehensive personal diabetes management platform.

Our next priorities include:

  • More sophisticated longitudinal tracking of medication responses and health indicators.
  • Better ways to identify meaningful changes and prepare users for doctor consultations.
  • Further clinical review of symptom pathways, thresholds, and educational content.
  • Expanded Bangla health education and accessibility.
  • More robust testing with real users and healthcare professionals.

Diabetes doesn't take a break after a doctor's appointment. It follows you to the dinner table, into your daily routine, and even into your sleep.

We want to make sure people don't have to navigate all of that alone.

Diabetes-: Because understanding your condition is just as important as treating it.

Built With

Share this project:

Updates

Submission history