Problem
London discards 1.9M tonnes of consumable food annually while 2M residents face food insecurity. Existing redistribution is fragmented with no real-time coordination.
Solution
AI-powered matching platform connecting food donors to charities with routing optimization. Designed complete technical architecture and user experience for both sides of the marketplace.
My Role
Product Designer & Technical Architecture
Tech Stack
- AI pattern recognition
- data stream analysis
- XGBoost
- ARIMA
- PostgreSQL
Main Project Documentation
📋
Main Project Report
5.8 MB • comprehensive report
Complete technical documentation covering AI architecture, implementation details, evaluation metrics, and future roadmap
Overview
Second Bite is an AI-powered platform design addressing food waste and food insecurity in London. The project transforms how surplus food redistribution operates by creating an intelligent marketplace that connects commercial food donors with charitable organizations in real-time.
Developed as part of Managing AI course at LSE, exploring ethical challenges in AI implementation and socio-technical system design.
Challenge Context
- Course: Managing AI (LSE)
- Brief: Design AI solution addressing social challenge with ethical implementation
- Focus: Algorithmic fairness, data governance, and multi-stakeholder systems
- Team: 4 members
Problem Analysis
- Scale: 1.9M tonnes of food waste vs. 2M food-insecure residents in London
- Market Gap: Existing solutions (OLIO, Too Good To Go) focus on consumer-level, not institutional redistribution
- Fragmentation: No centralized coordination between major donors (supermarkets) and receivers (charities)
- Efficiency: Manual matching processes waste time and food
Solution Design
Platform Architecture
Multi-sided marketplace connecting:
- Donors: Supermarkets, restaurants, food suppliers
- Receivers: Food banks, charities, community organizations
- Logistics: Route optimization for collection/delivery
Five-Stage Technical Pipeline
Stage 1: Data Collection
- API integration with inventory systems (SAP, Oracle)
- Manual entry with OCR validation for smaller businesses
- PostgreSQL database with real-time validation
Stage 2: Data Preprocessing
- Intelligent deduplication and quality validation
- Feature engineering: shelf life calculation, item categorization
- Geospatial features for distance and accessibility
Stage 3: Predictive Analytics
- ARIMA models: Forecast donor availability patterns
- XGBoost: Predict charity demand based on historical data
- Seasonal adjustment for holidays and events
Stage 4: Optimization & Allocation
- Linear programming for allocation decisions
- Multi-objective optimization: efficiency + fairness
- Vehicle routing problem (VRP) for delivery routes
- Bipartite graph matching for donor-charity pairing
Stage 5: Continuous Learning
- Reinforcement learning for strategy improvement
- Real-time feedback integration
- Performance metrics tracking
Key Features
- Real-time matching algorithm balancing efficiency and equity
- Predictive analytics for supply/demand forecasting
- Route optimization minimizing delivery time and cost
- Fairness constraints ensuring equitable distribution across communities
- Mobile-first interface for field operations
My Contributions
- Technical Architecture: Designed complete system architecture including data pipeline, ML models, and optimization engine
- Product Design: Created user journeys for both donors and receivers
- Ethical Framework: Developed algorithmic governance addressing bias and fairness
- Team Leadership: Coordinated 4-person team across research and design phases
Innovation Highlights
- Ethical AI Framework: Built fairness constraints directly into optimization algorithms
- Multi-objective Optimization: Balances efficiency metrics with equity considerations
- Regulatory Anticipation: Designed flexible architecture for evolving food safety regulations
- Community Focus: Algorithm prioritizes vulnerable communities using deprivation indices
Technical Challenges Addressed
- Real-time Matching at Scale: Hybrid optimization achieving sub-second matching times
- Data Quality: ML-based anomaly detection and automated cleaning
- Algorithmic Fairness: Multi-objective framework balancing efficiency and equity
- Regulatory Compliance: Modular architecture adapting to food safety regulations
User Experience Design
Donor Interface:
- Streamlined inventory input with OCR for receipts
- Real-time matching notifications
- Impact dashboard showing waste reduction
Receiver Interface:
- Needs profiling and capacity assessment
- Preference learning from feedback
- Nutritional analytics and beneficiary tracking
Impact Projections
- Environmental: 30% reduction in commercial food waste (570K tonnes diverted)
- Social: Serving 500K+ food-insecure individuals
- Economic: £45M in redistributed food value annually
- Operational: 95% successful delivery completion rate
Business Model
- Platform fees (2-3%) on successful matches
- Premium analytics for large donors
- Logistics revenue sharing
- Data insights for food industry stakeholders
Lessons Learned
- Stakeholder Complexity: Food redistribution involves competing interests requiring careful balance
- Ethical AI Design: Fairness and transparency must be built in from the start, not added later
- Real-world Constraints: Algorithmic sophistication must work within practical operational limits
- Social Impact Measurement: Quantifying social good requires sophisticated frameworks
Course Focus: Ethical AI Implementation
This project explored key challenges in Managing AI:
- Algorithmic bias detection and mitigation
- Explainable AI for stakeholder trust
- Data governance and privacy (GDPR compliance)
- Multi-stakeholder system design
- Balancing automation with human oversight
Additional Documentation
📊
Initial Pitch Presentation
1.2 MB • initial concept
Original project concept and problem identification for addressing food waste and insecurity in London