NourishIQ

AI-powered dining hall platform that cuts food waste 30-40% through demand forecasting, real-time waste tracking, and smart menu optimization—saving money and the planet, one meal at a time.

Comment

NourishIQ

AI-powered dining hall platform that cuts food waste 30-40% through demand forecasting, real-time waste tracking, and smart menu optimization—saving money and the planet, one meal at a time.

Team Members

  • John Song
  • Rishith Auluka
  • Hiep Pham

Purpose

University dining halls waste 30-40% of prepared food annually—resulting in tens of thousands of pounds of waste, hundreds of thousands of dollars in losses, and a massive carbon footprint. NourishIQ transforms this problem into an opportunity by providing dining managers with AI-powered tools to:

  • Predict demand with 7-day-ahead forecasting for every dish
  • Optimize menus using multi-objective AI scoring algorithms
  • Track waste in real-time with actionable insights
  • Measure impact through tangible sustainability metrics

A single dining hall implementing NourishIQ recommendations could save an estimated $945/month while preventing hundreds of kilograms of CO₂ emissions.

Tools & Technologies

Frontend Framework

  • Next.js 14 - React framework with App Router
  • TypeScript - Full type safety across 182+ type definitions
  • Tailwind CSS - Custom design system with sustainability-focused color palette

UI Libraries & Components

  • Recharts - Data visualization (charts, sparklines, heatmaps)
  • @hello-pangea/dnd - Drag-and-drop menu planner
  • Framer Motion - Smooth page transitions and animations
  • Lucide React - Icon system

AI & Data Processing

  • Custom algorithms built from scratch:
    • Demand forecasting engine
    • Waste risk classification system
    • Menu optimization multi-objective solver
    • Trend detection & alert generation
  • Seeded PRNG for reproducible 90-day synthetic dataset

Development Tools

  • ESLint - Code quality
  • Git - Version control

Challenges & Solutions

Challenge 1: Realistic AI Without External APIs

Problem: We wanted sophisticated AI predictions but couldn't rely on external APIs (cost, latency, API keys).

Solution: Built four custom AI modules from scratch using deterministic algorithms:

  • Demand Forecasting: Time-series pattern recognition with day-of-week weighting and event detection
  • Waste Classification: Composite risk scoring with weighted factors (waste %, refill frequency, depletion rate)
  • Menu Optimization: Multi-objective function (40% popularity + 40% efficiency + 20% nutrition) with constraint satisfaction
  • Alert Engine: Threshold-based anomaly detection with severity ranking

Challenge 2: Generating Believable Synthetic Data

Problem: Needed 90 days of realistic dining hall data that exhibited real-world patterns (exam weeks, weekend drops, meal-specific trends).

Solution: Implemented seeded pseudo-random number generation with:

  • Base attendance patterns (Friday +25%, Monday -18%)
  • Event modifiers (exam week +35% comfort food)
  • Meal-specific waste profiles (late lunch higher waste)
  • Seasonal drift and noise injection
  • Result: Reproducible, statistically coherent dataset

Challenge 3: Real-Time Interactive Menu Planner

Problem: Drag-and-drop interface needed to update AI recommendations instantly based on selected time slot.

Solution:

  • Used @hello-pangea/dnd for smooth drag interactions
  • Implemented context-aware AI sidebar that recalculates scores on slot selection
  • Optimized re-rendering with React memoization
  • Added live summary bar that updates with every menu change

Challenge 4: Making Sustainability Tangible

Problem: "Reduced 150 kg of waste" doesn't emotionally resonate with users.

Solution: Translated metrics into relatable equivalents:

  • CO₂ saved → trees planted (industry standard: 2.5 kg CO₂/kg waste)
  • Water saved → showers (1,500 L/kg waste)
  • Created visual impact reports with these conversions
  • Result: Sustainability becomes personal and memorable

Challenge 5: Managing Complex State Across Pages

Problem: Multiple pages needed to share filters, date ranges, and AI configurations without prop drilling.

Solution:

  • Designed modular component architecture
  • Used TypeScript interfaces for strict type contracts
  • Implemented URL state management for shareable analytics views
  • Separated data layer from presentation layer

How We Built NourishIQ

1. Research & Discovery

We interviewed dining hall managers and researched industry waste patterns to understand:

  • Common pain points (overproduction, lack of visibility, manual planning)
  • Key performance indicators (waste %, cost per meal, sustainability score)
  • Decision-making workflows (daily prep adjustments, weekly menu planning)

2. Architecture Design

Designed a 7-page application with clear information hierarchy:

  • Dashboard: High-level KPIs and AI recommendations
  • Predictions: 7-day demand forecasts with confidence scores
  • Analytics: Deep-dive dish performance with filters
  • Menu Planner: Drag-and-drop AI-assisted scheduling
  • Reports: PDF-exportable sustainability audits
  • Produce Tracking: Fresh item uptake monitoring
  • Settings: Customization and alert configuration

3. AI Algorithm Development

Built the core intelligence layer:

Demand Forecasting Model: $$\text{Predicted Demand} = \text{Base Avg} \times \text{Day Factor} \times \text{Event Factor} \times (1 + \text{Trend})$$

Where:

  • $\text{Day Factor} \in [0.82, 1.25]$ based on historical day-of-week patterns
  • $\text{Event Factor}$ detects exam weeks, holidays, weather
  • $\text{Trend}$ uses 7-day moving average slope

Confidence Score: $$\text{Confidence} = \max\left(65, \min\left(95, 100 - \left(\frac{\sigma}{\mu} \times 100\right)\right)\right)$$

Based on coefficient of variation from historical data.

Waste Risk Score: $$\text{Risk} = 0.4 \times W_{\%} + 0.3 \times R_{\text{freq}} + 0.2 \times (1 - D_{\text{rate}}) + 0.1 \times T_{\Delta}$$

Where $W_{\%}$ = waste percentage, $R_{\text{freq}}$ = refill frequency, $D_{\text{rate}}$ = depletion rate, $T_{\Delta}$ = trend change.

Menu Optimization Score: $$\text{AI Score} = 0.4 \times P_{\text{pop}} + 0.4 \times E_{\text{eff}} + 0.2 \times N_{\text{bal}}$$

Constraints: Budget limit, category diversity, nutritional minimums.

4. UI/UX Implementation

  • Built custom heatmap component with color-coded waste severity
  • Designed status badge system (All-Star, Improving, Watch, Red Flag)
  • Implemented 14-day sparkline trends for quick pattern recognition
  • Created responsive layouts tested across devices

5. Data Visualization

Integrated Recharts for:

  • Area charts (waste trends over time)
  • Bar charts (comparative metrics)
  • Composed charts (multi-metric overlays)
  • Pie charts (waste distribution by category)
  • Custom tooltips with contextual information

6. Testing & Refinement

  • Validated AI predictions against synthetic patterns
  • User-tested drag-and-drop interactions for intuitiveness
  • Optimized load times (lazy loading, code splitting)
  • Ensured accessibility (keyboard navigation, ARIA labels)

What We Learned

Technical Learnings

  • TypeScript mastery: Managing 182 type definitions taught us the value of strict typing for large applications
  • Algorithm design: Building AI from scratch deepened our understanding of forecasting, optimization, and classification
  • State management: Learned when to use local state vs. context vs. URL parameters
  • Performance optimization: Discovered re-rendering bottlenecks and fixed them with memoization

Domain Learnings

  • Sustainability metrics: Learned industry standards for carbon footprint calculations
  • Food service operations: Discovered the complexity of dining hall logistics (staffing, prep times, ingredient ordering)
  • Behavioral patterns: Understood how student behavior affects demand (exam weeks, Fridays, weather)

Design Learnings

  • Data storytelling: Numbers alone don't persuade—context and relatable equivalents do
  • Progressive disclosure: Complex systems need layered interfaces (overview → details → actions)
  • Color psychology: Green = sustainability, red = alerts, blue = neutral data

Teamwork Learnings

  • Version control discipline: Consistent commit messages and branch strategies prevented merge conflicts
  • Code reviews: Caught bugs early and shared knowledge across the team
  • Documentation: Writing README and inline comments clarified our own thinking

What Inspired Us

The inspiration came from three converging realizations:

  1. Environmental urgency: Food waste is the 3rd largest contributor to greenhouse gas emissions globally. If food waste were a country, it would be the third-largest emitter after the US and China.

  2. Local impact: Many of us witnessed dining hall waste firsthand—full trays discarded, overflowing bins, staff preparing far more than needed. The problem was visible but unsolved.

  3. AI opportunity: Recent advances in demand forecasting and optimization are used in retail and logistics, but rarely in institutional dining. We saw a gap where AI could create immediate, measurable impact.

We wanted to build something that combined technical sophistication with real-world utility—a system that could actually be deployed, not just demoed.

Key Features

1. Real-Time Dashboard

  • 4 hero KPIs with trend indicators
  • AI-generated waste heatmap (station × meal period)
  • Prioritized recommendation engine with confidence scores
  • One-click accept/dismiss actions

2. 7-Day Demand Forecasting

  • Dish-level predictions for every meal
  • 65-95% confidence scoring
  • Event detection (exam weeks, holidays)
  • Historical baseline comparison

3. Waste Alert System

  • 8 active alerts ranked by severity
  • Root cause analysis (why waste is happening)
  • Specific recommendations with dollar impact
  • Trend tracking (improving/deteriorating)

4. AI-Assisted Menu Planner

  • Drag-and-drop weekly calendar interface
  • Context-aware dish recommendations per time slot
  • Live optimization score updates
  • Multi-objective AI scoring (popularity + efficiency + nutrition)

5. Impact Reporting

  • Weekly summaries, monthly reports, sustainability audits
  • CO₂ → trees conversion (2.5 kg CO₂/kg waste)
  • Water → showers conversion (1,500 L/kg waste)
  • PDF export functionality

6. Produce Uptake Tracking

  • 9-item fresh produce monitoring
  • 7-day sparkline trends
  • Uptake percentage with AI swap recommendations
  • Data-driven phase-in/phase-out decisions

Impact Potential

Based on industry benchmarks and our AI models:

Metric Current State With NourishIQ Improvement
Waste Rate 30-40% 18-25% ~40% reduction
Monthly Savings — $945/hall $11,340/year
CO₂ Prevented — 1,200 kg/month ~6 trees/month
Water Saved — 180,000 L/month ~1,200 showers

For a university with 5 dining halls, annual impact:

  • $56,700 saved
  • 7.2 tons CO₂ prevented
  • 360 tree-equivalents

Future Enhancements

  1. Machine Learning Integration: Replace deterministic algorithms with trained models (LSTM for time series, gradient boosting for classification)
  2. IoT Integration: Connect to smart scales and RFID systems for automatic waste tracking
  3. Mobile App: Staff input interface for real-time adjustments
  4. Multi-Location Support: District-level dashboard for university system oversight
  5. Recipe Database: Integrate with nutrition APIs for automated nutritional scoring
  6. Student Feedback Loop: Dish rating system to improve popularity predictions

Credits & Frameworks

This project was built with the following open-source technologies:

Core Frameworks

UI Libraries

Development Tools

Algorithms & Concepts

  • Demand forecasting methodology inspired by retail inventory management research
  • Carbon footprint calculations based on EPA food waste standards
  • Multi-objective optimization concepts from operations research literature

All AI algorithms were developed in-house. No external AI APIs were used.

Acknowledgments

We'd like to thank:

  • Dining hall managers who shared insights during our research phase
  • Sustainability officers who helped us understand carbon accounting
  • The open-source community for the incredible tools that made this possible

Built for a sustainable future

Built With

Share this project:

Updates