Eat Smart AI (Application-Building Task)

Dashboard Page in dark theme
Dashboard Page in light theme
New chat
Click view chat history
Viewing chat history
Viewing generated meal planner part 1
Viewing generated meal planner part 2
Upload nutrition facts image and description
Generated results will show with calories, fats, carbs and so on.
User able to view recent analyzed meal
User able to add meal to tracker
Add to tracker successfully
User saved meal as favourite
User view diet tracker
User able to add meals by selecting analyzer meals
User able to add meals by manually inputting meals details
User able to view own daily summaries
User able to view last 30 days' daily summaries
User able to view own water intake
User able to view own profile
User able to add their allergies and restrictions and food to avoid
User able to customize their app settings

Inspiration

Many existing nutrition applications focus primarily on calorie counting, which often feels repetitive and difficult to sustain in the long term. While users may know how much they eat, they rarely understand what they are eating or how it impacts their health. We wanted to build a smarter and more supportive system that interprets food in a more human-like way, provides meaningful nutritional insights, and adapts to different dietary habits and cultures.

The ERNIE Hackathon inspired us to explore how large language models can be applied to real-world health challenges and improve everyday decision-making through AI.

What It Does

Tracks daily meals, water intake, and macronutrients in a single platform
Uses AI-based image and text analysis to identify food and estimate nutritional values
Generates personalized 7-day meal plans based on user goals and preferences
Provides a smart nutrition chatbot for guidance and explanations
Displays daily summaries, nutrition trends, and diet quality scores
Supports both English and Mandarin to improve accessibility

How We Built It

Frontend

Framework: Next.js 15 (App Router) with TypeScript
Styling: Tailwind CSS v4, shadcn/ui, and Radix UI
Data Visualization: Recharts for charts and analytics

Backend and Data

Supabase
- User authentication
- PostgreSQL database
API Routes
- Secure communication between the frontend and AI services

Artificial Intelligence

LLM Integration
- Baidu Ernie 5.0 Thinking
- Baidu Ernie 4.0-8K (latest)
- Integrated via the Baidu AI Studio LLM API
AI Client Configuration
- Custom OpenAI-compatible client
- Custom baseURL pointing to Baidu AI Studio
- Secure server-side API key handling
Prompt Engineering
- Food image analysis
- Nutrition estimation
- Meal planning
- Conversational chatbot responses

Deployment

Platform: Vercel

Version Control

Git and GitHub

System Design

Modular and scalable architecture
Optimized for performance and user experience

Challenges We Ran Into

Ensuring reasonable nutrition estimation accuracy from images and text
Designing prompts that generate structured and consistent AI responses
Managing multilingual support while preserving conversation context
Balancing real-time AI responses with smooth UI performance
Coordinating multiple AI-powered features within a single application

Accomplishments That We’re Proud Of

Successfully applying ERNIE 4.0 and 5.0 across multiple features in a real-world application
Delivering a fully functional, end-to-end AI-powered web platform
Creating an intuitive dashboard that simplifies nutrition tracking
Implementing bilingual AI interaction with persistent chat history
Deploying a live demo that demonstrates practical AI usage beyond a prototype