Transform Your Health with AI-Powered Food Therapy

Get personalized dietary recommendations based on Traditional Chinese Medicine principles—all in just a few clicks

Challenge

Low Accuracy due to image segmentation and model optimization.

Solution

Limiting to 4 tongue types can improve accuracy in segmentation and model optimization.

Enhance Generalization: helps the model recognize different features and reduces overfitting.
Improve Segmentation: allows a better understanding of tongue variations, leading to more accurate segmentation.
Optimize Training: Focusing on distinct categories helps the model learn specific characteristics, improving overall accuracy.

My Role

Project Time

Product Designer

Aug 2024 - Nov 2024

Tools

Team of 8

2 Product Manager, 3 Developers, 2 Product Designers

Figma, Google Meet, Discord, Trello

Design Outcome

We identified four tongue types for better accuracy in the initial version. Each tongue type includes an analysis profile and food and herbal tea recommendations.

Aim to improve user engagement through a visually appealing UI, useful quiz content, and personalized food and tea suggestions based on results.
This enhances user satisfaction and encourages continued interaction with the platform. Clear instructions for uploading tongue photos are also essential for a smooth user experience.

My challenge for this project

Design Process

Overview

Current health tools lack traditional Chinese medicine techniques. This app aims to bridge this gap by introducing tongue photo analysis for holistic health assessment. This is an AI tool that helps improve health through better eating. Based on Traditional Chinese Medicine and natural health, our app uses tongue image recognition and data analysis to provide tailored diet advice.

Persona

  Busy Professional - Sarah

Health-Conscious Enthusiast - Jason

Parent and Caregiver - Maria

Elderly Health Seeker - Robert

Unmet needs

personalized health and dietary guidance tailored to individual preferences, health goals, and lifestyles
require convenient, time-saving solutions such as quick meal plans, simple recipes, or ready-to-eat healthy options
engagement and motivation tools are essential to keep them on track, including reminders, progress tracking, and interactive features that promote consistency and accountability

Solutions

need quick and easy health analysis to help users understand their current well-being
simple and personalized dietary recommendations based on Traditional Chinese Medicine (TCM) principles, tailored to individual needs
Continuous engagement is fostered through sharing features and regular updates in our community blog, offering fresh content to keep users motivated and informed

Emotional Journey

Curiosity: Users are intrigued by the app’s unique TCM-based health insights using tongue analysis.
Trust: As they receive tailored diet advice grounded in TCM, users begin to trust the app’s holistic approach.
Satisfaction: Experiencing positive health changes, users feel satisfied and engaged.
Loyalty: The app becomes a trusted wellness companion, fostering lasting loyalty through culturally resonant support

User Onboarding Journey

Initial Assessment

Users take and upload a tongue photo for baseline analysis

Health Data Input

Users input key health metrics for a comprehensive profile

Core Functionalities

Personalized Feedback & Food Recommendation

The app provides immediate, tailored dietary and lifestyle advice
Users easily track option of their meals for real-time analysis

Image Recognition: AI analyses tongue to assess health indicators.
TCM Integration: Data is interpreted using Traditional Chinese Medicine principles.
Personalized Recommendations: Tailored dietary advice based on holistic health analysis.

User Flow

The video demonstrates the two types of user flows:

User Flow 1 - Analysis result without food recommendation

Photo Consent: User agrees to upload a tongue photo. —> Photo Upload: User uploads or takes a tongue photo. —> Skip Questionnaire: Option to bypass the questionnaire. —> Analysis Result: Display personalized analysis based on the photo.

User Flow 2 - Analysis result with food recommendation

Photo Consent: User agrees to upload a tongue photo. —> Photo Upload: User uploads or takes a tongue photo. —> Answer Questions: User completes 7 short questions. —> Analysis Result: Display personalized health analysis. —> Food Recommendations: Provide tailored dietary suggestions based on the analysis.

Prototype

Future Enhancements

Core Technology Overview

Frontend:
- ReactJS - A JavaScript library for building the user interface of the web application, deployed on AWS EC2.
Backend APIs:
- Python/Flask for handling image classification requests, deployed on AWS EC2.
- API Gateway + Lambda (Python) for handling questionnaire submissions and routing requests.
Data Storage:
- Amazon DynamoDB - A fully managed NoSQL database service for storing user data, questionnaire responses, and analysis results.
- Amazon S3 - A scalable object storage service for temporarily storing uploaded photos for analysis and other necessary files.
ML Model
- ResNet 34 - A convolutional neural network (CNN) model for image classification, fine-tuned using PyTorch in Jupyter Notebook, and deployed on AWS EC2.
Model Fine-tuning
- Jupyter Notebook - An interactive web application for developing and fine-tuning the machine learning model.
Deployment Infrastructure

Amazon EC2: Used for deploying the frontend (ReactJS), backend API (Flask), and the image classification model (ResNet 34).
API Gateway + Lambda: Manages serverless processing of questionnaire submissions.