Overview
A comprehensive health management platform integrating LINE Bot, AI image recognition, and personalized health coaching to help users track their diet and receive tailored health advice.
Project Duration
March 2025 - June 2025
My Role & Contributions
AI Integration
- Integrated OpenAI GPT-4o/GPT-4o-mini for conversational AI
- Implemented Google Gemini for food image recognition
- Developed nutrition analysis pipeline
LINE Bot Development
- Built LINE Messaging API integration
- Implemented AI-powered health conversations
- Created message push notification system
Backend Architecture
- Designed and implemented RESTful APIs
- Built Hangfire scheduled task system
- Developed admin dashboard backend services
- Refactored image processing pipeline using SkiaSharp
Admin Dashboard
- User management interface
- Diet record tracking
- Questionnaire management
- Reward system administration
Tech Stack
Backend
- Framework: ASP.NET Core 6
- ORM: Entity Framework Core
- Database: SQL Server
- Job Scheduler: Hangfire
- Image Processing: SkiaSharp
Frontend
- Framework: Vue 3
- UI Library: Vuetify 3
- Language: TypeScript
AI Services
- Conversation: OpenAI GPT-4o / GPT-4o-mini
- Image Recognition: Google Gemini
- Analysis: Custom nutrition algorithms
Integration
- Messaging: LINE Messaging API
- Push Notifications: LINE Push API
Key Features
For Users
- Food photo analysis with AI
- Nutritional breakdown and recommendations
- Daily diet tracking
- Personalized health advice
- Scheduled reminders and tips
For Administrators
- User management dashboard
- Diet record monitoring
- Questionnaire creation and analysis
- Reward system management
- System analytics and reports
Technical Highlights
Image Processing Refactoring
Refactored the entire image processing workflow using SkiaSharp to improve:
- Cross-platform compatibility
- Performance optimization
- Memory management
- Image quality
Scheduled Tasks
Implemented robust scheduling system with Hangfire:
- Daily health reminders
- Nutrition report generation
- Automated message dispatch
- Data cleanup tasks
AI Integration Strategy
- Fallback mechanisms between AI providers
- Cost optimization through model selection
- Response caching for common queries
- Rate limiting and quota management
Challenges & Solutions
Challenge: AI response consistency Solution: Implemented prompt engineering techniques and response validation
Challenge: High volume image processing Solution: Asynchronous processing queue with SkiaSharp optimization
Challenge: Real-time LINE webhook handling Solution: Event-driven architecture with message queuing
Results
- Successfully deployed to production
- Handled 10,000+ food image analyses
- 95%+ user satisfaction rate
- Reduced manual nutritionist workload by 60%