sweeten

• 

  landing page

About the project

Inspiration

Sweeten was inspired by both personal experience and a larger public health problem. My mother was diagnosed with type 2 diabetes, and I observed how challenging it was for her to manage daily glucose readings, diet, and activity without clear guidance. At the same time, in Pakistan, diabetes affects more than 34 million adults — roughly one in three people aged 20–79 (WHO). Millions face similar daily management challenges, often with limited access to healthcare. This combination of personal and societal motivation led to the idea for Sweeten.

What it does

Sweeten transforms daily health data entered by users into weekly, personalized diabetes management plans. It uses a rule-based core system to generate guidance, with an optional AI layer to refine phrasing and personalization. The plans include suggestions for diet, physical activity, and behavioral adjustments. The system works fully on its rules and can be checked and understood even when AI or internet access is limited.

How I built it

Sweeten is a modular web application built with Next.js 15, TypeScript, and Tailwind CSS. User data is stored in Firebase Firestore, with Authentication managing access. The plan engine follows a deterministic, rules-first design:

1. Data collection – Users input daily vitals via a calendar interface.
   
2. Plan generation – The system classifies glucose trends, activity, and other parameters to select a weekly focus plan.
   
3. Optional AI refinement – Google Gemini adjusts tone and personalization without altering safety-critical logic.
   

This hybrid approach ensures the system produces reliable plans even under limited resources.

Challenges I ran into

• Translating clinical guidelines into clear, deterministic rules that remain safe and actionable.
  
• Handling missing or inconsistent user data while keeping plans meaningful.
  
• Designing a hybrid system where AI enhances personalization without affecting core safety logic.
  
• Building a clean, responsive interface that works across devices and maintains clarity.

Accomplishments that I'm proud of

• Developed a fully functional rule-based plan engine with optional AI refinement.
  
• Designed a system that works reliably even with partial data or offline scenarios.
  
• Linked personal experience to a real-world public health problem, showing societal relevance.
  
• Created a transparent and auditable system suitable for competitions.

What I learned

• How to implement modular front-end and back-end architecture for health apps.
  
• How to structure health data for longitudinal analysis in Firestore.
  
• Practical insights into AI-human hybrid systems, emphasizing safety, transparency, and reliability.
  
• The importance of clear communication and usability for non-expert users managing chronic conditions.

What's next for Sweeten

• Conduct clinician-validated evaluations to refine guidance.
  
• Adapt guidelines specifically for the Pakistani context.
  
• Integrate with wearables and glucometers for automated data collection.
  
• Explore sustainable deployment through partnerships with hospitals or public health organizations.
  
• Expand the system to provide long-term trend tracking and habit reinforcement features.

Detailed Presentation Doc

Built With

• firebase-(authentication-+-firestore)
• framer-motion
• google-gemini-(ai-refinement-via-api)
• next.js-15
• react-icons
• tailwind-css
• typescript