menu
Main page
gemini prompting example
checkout page
admin page (static for now)
About page
kitchen dashboard to see orders come in

TerpEats - Order Smart. Eat Fresh. Waste Less.

Inspiration

As University of Maryland students, we witnessed firsthand the daily struggles of campus dining: long lines at peak hours, uncertainty about menu options, and the challenge of maintaining nutritional goals while juggling a busy academic schedule. We saw students either skipping meals or settling for less healthy options simply because they didn't have time to plan ahead.

The inspiration for TerpEats came from a simple observation: what if students could order ahead, see exactly what's available, and get personalized nutrition guidance—all in one place? We wanted to create a solution that not only saves time but also promotes healthier eating habits and reduces food waste by enabling better demand forecasting for dining halls.

What It Does

TerpEats is a comprehensive campus dining platform that revolutionizes how students interact with dining halls:

Core Features:

Pre-Order System: Students can browse menus from all three UMD dining halls (South Campus, Yahentamitsi, 251 North) and place orders in advance
Real-Time Menu: Access to 3,149 menu items with detailed nutritional information (calories, protein, carbs, fat)
Smart Filtering: Filter by dining hall, meal period (breakfast, brunch, lunch, dinner), and search by item name
AI Nutrition Coach: Gemini-powered chatbot that provides personalized meal recommendations based on your profile (age, height, weight, activity level, and macro goals)
Order Tracking: Real-time order status updates from "pending" → "in progress" → "ready" → "picked up"
Kitchen Dashboard: Live view for dining hall staff to manage incoming orders
Nutrition Tracking: Automatic daily nutrition logging to help students meet their macro goals
Profile Management: Set and track personal fitness goals with BMR/TDEE calculations

The Math Behind It

We use the Mifflin-St Jeor Equation to calculate Basal Metabolic Rate (BMR):

For Males: $$BMR = 10 \times \text{weight}{kg} + 6.25 \times \text{height}{cm} - 5 \times \text{age} + 5$$

For Females: $$BMR = 10 \times \text{weight}{kg} + 6.25 \times \text{height}{cm} - 5 \times \text{age} - 161$$

Then we calculate Total Daily Energy Expenditure (TDEE): $$TDEE = BMR \times \text{Activity Factor}$$

Where Activity Factor ranges from 1.2 (sedentary) to 1.9 (very active).

How We Built It

Tech Stack:

Frontend:

React with Vite for fast development and optimal performance
Tailwind CSS for beautiful, responsive UI with UMD branding (red and gold theme)
React Router for seamless navigation
Context API for state management (Auth, Cart)

Backend & Database:

Supabase (PostgreSQL) for real-time database with Row Level Security
Supabase Auth for secure user authentication
Real-time subscriptions for live order updates in the kitchen dashboard

AI Integration:

Google Gemini 2.5 Flash for personalized nutrition coaching
Express.js proxy server to handle Gemini API calls server-side (avoiding CORS and securing API keys)

Data Pipeline:

Scraped 3,149 menu items from UMD dining halls
CSV processing and normalization
Dual data source: Supabase (primary) + CSV fallback for reliability

Architecture Highlights:

Optimized Queries: Limited Supabase queries to 500 items with selective column fetching to prevent resource exhaustion
Fallback System: 15-second timeout with automatic CSV fallback ensures the app always works
Real-time Updates: Kitchen dashboard uses Supabase channels for instant order notifications
Smart Deduplication: Normalized menu items to remove duplicates while preserving meal-period variations
Imperial Units: User-friendly input (feet/inches, pounds) with automatic metric conversion for calculations

Challenges We Faced

1. Supabase Resource Limits

Problem: ERR_INSUFFICIENT_RESOURCES when querying 3,149 menu items Solution:

Reduced query to essential columns only
Implemented pagination with .limit(500)
Created CSV fallback system for reliability
Added 15-second timeout with graceful degradation

2. Gemini API Integration

Problem: Browser CORS restrictions and model endpoint compatibility Solution:

Built Express.js proxy server to handle API calls server-side
Upgraded to @google/generative-ai v0.2.1 for v1 endpoint compatibility
Implemented fallback from gemini-1.5-flash to gemini-2.5-flash
Added markdown stripping (**text** → text) for cleaner chat responses

3. Data Normalization

Problem: Inconsistent nutrition data formats (e.g., "total carbohydrate. 0.5g") Solution:

const toNumber = (value) => {
  if (typeof value === 'number') return value
  const match = String(value).match(/-?\d+(?:\.\d+)?/)
  return match ? parseFloat(match[0]) : null
}

4. Order Items Foreign Key Constraint

Problem: Menu items from CSV didn't exist in Supabase menu_items table Solution:

Modified order_items schema to store item details directly (name, calories, macros)
Made menu_item_id nullable
Orders now work independently of menu item database entries

5. React Import Errors

Problem: "React is not defined" errors in JSX files Solution: Added explicit import React from 'react' to all component files

6. Dining Hall Dropdown Loading

Problem: Dropdown stuck on "Loading dining halls..." Solution: Hardcoded the three UMD dining halls instead of fetching from database for instant availability

7. Cart Clearing Flow

Problem: Empty cart screen showing after order placement Solution: Reordered operations to clear cart → show modal → navigate on button click (not automatically)

Use of AI in Development

We intentionally leveraged AI tools including ChatGPT, Google Gemini, and Cursor AI to enhance our development workflow and code quality throughout the project.

AI was used to:

Debug complex React and Supabase integration issues
Suggest optimal UI layouts and component architecture patterns
Optimize SQL queries and database schema design
Generate and refine Tailwind CSS styling
Clean and organize project documentation
Improve nutrition calculation logic and BMR/TDEE formulas
Speed up repetitive coding tasks and boilerplate generation
Parse and normalize inconsistent CSV data formats
Troubleshoot CORS and API integration challenges

Our Approach:

Every AI-generated code snippet was carefully reviewed, tested, and manually integrated into our codebase. We used AI as a coding assistant to accelerate development and explore solutions, not as a substitute for our own engineering decisions. All architectural choices, feature implementations, and problem-solving strategies were driven by our team's understanding and requirements.

What We Learned

Real-time Systems: Implementing Supabase real-time subscriptions taught us about WebSocket-based architectures and the importance of efficient database queries
API Proxy Patterns: Building a server-side proxy for Gemini API showed us how to handle CORS, secure API keys, and provide fallback mechanisms
Data Resilience: The importance of fallback systems—our CSV backup ensures the app works even when Supabase has issues
User Experience: Small details matter—imperial units, 15-second timeouts, clear error messages, and confirmation screens all contribute to a polished experience
Performance Optimization: Query optimization, selective column fetching, and pagination are crucial for working within free-tier limits
Nutrition Science: Implementing BMR/TDEE calculations and macro tracking gave us deeper insights into nutrition and fitness

What's Next for TerpEats

Push Notifications: Alert students when their order is ready
Meal Planning: AI-powered weekly meal plans based on class schedules
Social Features: Share favorite meals and dining hall reviews
Analytics Dashboard: Insights for dining hall managers on popular items and peak hours
Dietary Restrictions: Advanced filtering for allergies, vegetarian, vegan, halal, kosher
Integration with UMD ID: Seamless payment with dining dollars
Waste Reduction: Predictive analytics to help dining halls optimize food preparation
Nutrition Goals Gamification: Streaks, achievements, and challenges for hitting macro goals