DineSync

Inspiration

"Where should we eat?" It's the question that has sparked countless debates, delayed dinner plans, and left someone in the group unhappy with the final choice.

We've all been there: endless group chat messages, conflicting preferences, dietary restrictions, budget constraints, and decision fatigue. Someone always ends up compromising too much, and the person who suggested the restaurant feels guilty if others don't enjoy it.

Meanwhile, Yelp's AI technology was revolutionizing how we discover restaurants. We asked ourselves: "What if AI could solve group dining decisions by ensuring everyone's preferences matter equally?"

That's how DineSync was born.

What it does

DineSync is an AI-powered group dining consensus platform that transforms the chaotic "where should we eat?" debate into a fair, transparent, and fast decision-making process.

Core Workflow:

Create Session - Host enters dining location and gets a shareable link
Collect Preferences - Each person submits cuisines, dietary needs, budget, distance, and ambiance preferences
Find Restaurants - Yelp AI searches for restaurants matching all criteria
Rank by Fairness - Custom algorithm ensures no one is left behind (70% average + 30% minimum satisfaction)
Visual Feedback - Individual satisfaction meters (0-100%) show who's compromising
AI Conversation - Ask questions about each restaurant to make informed decisions

Professional Features:

Real-Time Collaboration - Auto-refresh shows when friends submit preferences (5-second polling)
Fairness-Weighted Algorithm - 70/30 split maximizes group happiness while protecting individual satisfaction
Satisfaction Meters - Visual transparency for each person's happiness level
Dietary Safety - Hard enforcement of restrictions (violation = 0% score)
Veto Power - Absolute dealbreakers for allergies or strong dislikes
AI Conversation Mode - Context-aware answers about location, ambiance, dietary options
Compromise Indicators - Shows who's making the biggest sacrifice
Top 10 Results - Ranked by group satisfaction with detailed breakdowns
Multi-Cuisine Support - 2-3 restaurant variations per selected cuisine
Booking Integration - Reserve tables through Yelp AI (simulated)

How we built it

Tech Stack:

Frontend: Streamlit 1.37+ (Interactive multi-page web UI)
Backend: Python 3.11 (Application logic & services)
AI Engine: Yelp AI Chat API (Restaurant search & intelligent matching)
Consensus: Custom Fairness Algorithm (70/30 weighted scoring)
Validation: Pydantic 2.8+ (Data validation & models)
Session Management: JSON Files (Multi-user coordination)
Package Manager: uv (Fast dependency management)
Deployment: Streamlit Cloud (Production hosting)

Architecture:

We built a modular, production-ready system with clear separation of concerns:

Service Layer:
- yelp_ai_client.py - API communication with Yelp AI
- consensus_engine.py - Fairness-weighted group scoring algorithm
- session_manager.py - Multi-user session coordination
- performance_metrics.py - API performance tracking & monitoring
Data Layer:
- config.py - Environment configuration & settings
- sessions.json - Active sessions with participant data
- Pydantic schemas for type safety and validation
Application Layer:
- dinesync_app.py - Main Streamlit app (900+ lines)
- Multi-page flow: Create → Join → Results
- Real-time state management with auto-refresh

Development Process: Session management → Preference collection → Yelp AI integration → Consensus algorithm → Visual feedback → AI conversation mode → Production deployment

Challenges we ran into

1. Group Consensus Algorithm Design

Challenge: How to balance overall group happiness with individual satisfaction?
Solution: Researched voting theory and fairness algorithms, settled on 70/30 weighted scoring
Result: Prevents "tyranny of the majority" while maximizing overall satisfaction

2. Multi-User Session Coordination

Challenge: Streamlit is stateless; how to coordinate multiple users in real-time?
Solution: JSON-based session storage with auto-refresh polling every 5 seconds
Result: Seamless real-time collaboration without WebSockets or database

3. Dietary Restriction Enforcement

Challenge: Yelp AI doesn't always return dietary information consistently
Solution: Hard requirement validation - any violation results in 0% score
Result: Safe recommendations that respect all dietary needs

4. Yelp AI Query Optimization

Challenge: How to query for multiple cuisines without overwhelming results?
Solution: Strategic OR queries with 2-3 variations per cuisine, explicit dietary requirements
Result: Diverse restaurant options (6-15 results) that match group preferences

5. Satisfaction Score Transparency

Challenge: Users need to understand why a restaurant was ranked higher/lower
Solution: Individual score breakdowns with visual meters and compromise indicators
Result: Transparent decision-making that builds trust in recommendations

6. AI Conversation Context

Challenge: Generic AI responses don't help users make decisions
Solution: Context-aware prompts with restaurant details, location, and user preferences
Result: Detailed, actionable answers about ambiance, dietary options, and recommendations

Accomplishments that we're proud of

Technical Excellence:

✅ Fully functional, production-ready application deployed live
✅ Clean, modular architecture with clear separation of concerns
✅ Novel fairness-weighted consensus algorithm (70/30 split)
✅ Real-time multi-user coordination without database
✅ Robust error handling and edge case management
✅ Intelligent prompt engineering for consistent AI responses
✅ Zero external database dependency (JSON-based persistence)

Product Innovation:

✅ Solves universal problem that affects millions of dining decisions daily
✅ First app to use Yelp AI for multi-user group consensus
✅ Fairness algorithm ensures no one is left behind
✅ Visual satisfaction meters provide transparency
✅ AI conversation mode for informed decision-making
✅ Veto power for absolute dealbreakers
✅ Compromise indicators show who's sacrificing

User Experience:

✅ Intuitive 3-page flow anyone can use
✅ Shareable links for seamless group coordination
✅ Auto-refresh shows real-time participant updates
✅ Expandable results with top 3 expanded by default
✅ Mobile responsive design works on all devices
✅ Immediate feedback with clear visual indicators

Business Value:

✅ Addresses universal pain point in social dining
✅ Viral potential through shareable links and network effects
✅ Solves recurring problem (not one-time use)
✅ Scalable to any location with Yelp coverage
✅ Reduces decision fatigue and social friction

What we learned

Technical Insights:

Fairness Algorithms Matter: Simple weighted scoring (70/30) outperforms complex voting systems
State Management is Critical: Proper session state handling enables real-time collaboration
Prompt Engineering for Diversity: Explicit "2-3 variations per cuisine" prevents repetitive results
Hard Requirements Work: Zero-tolerance for dietary violations ensures safety
Visual Feedback Builds Trust: Satisfaction meters make algorithm decisions transparent

Product Lessons:

Solve One Problem Well: Better to excel at group consensus than add mediocre features
Transparency Drives Adoption: Users trust recommendations when they see individual scores
Real-Time Matters: Auto-refresh creates sense of collaboration even without WebSockets
Veto Power is Essential: Absolute dealbreakers (allergies) need special handling
AI Conversation Adds Value: Context-aware answers help users make informed decisions

Business Learnings:

Universal Pain Point: Everyone has experienced the "where should we eat?" debate
Network Effects: More users in a session = more value for everyone
Viral Potential: Shareable links encourage organic adoption
Recurring Use Case: Groups dine together repeatedly, creating retention
Market Validation: Friends immediately understood and wanted to use it