Designing a Dating App (Tinder/Bumble) - Mock Interview Notes

Session Overview

• Interviewer: Unnamed
• Interviewee: Yen Lee, Senior Software Engineer at TikTok
• Objective: Design a dating app similar to Tinder or Bumble

Key Functionalities & Requirements

1. Profile Creation
   • Bio setup
   • Preferences
   • Photos/Videos
   • Optional: Super like features
   • Important Components:
     • Images/Videos upload
     • Bio information
     • Preferences and filters
2. Feed/Recommendation
   • Local user recommendations
   • Active user prioritization
   • Profile matching algorithms
3. **Matching
   • Mechanism for users to match (e.g., swipe right)
   • Notification when matched
4. Messaging
   • Private messaging after matching
   • Exchanging texts, photos, and videos
5. Additional Considerations
   • System monitoring & logging (optional due to scope)
   • Subscription model for monetization

Traffic & Storage Estimations

• Estimated Users: 50 million
• Active Users: 1 million per day
• New Profiles: 500k per day
• Matches: 1 billion per day
• Storage Requirement:
  • Profile Photos: ~300-600TB
  • Messaging Data:
    • 200 million messages daily
    • 10 million photos uploaded daily

Detailed Design Components

Profile Creation Process

• Client Perspective: Flow
  • Users answer questions (JSON format)
  • Upload images/videos
  • Store data in NoSQL database
  • Encode and resize images for storage efficiency
• Profile Data Format:
  • Profile ID (string)
  • User ID (string)
  • Preferences (JSON)
  • Bio information
  • Photo IDs (list of strings)
  • Last Login information for activity status

Feed/Recommendation System

• Requirements:
  • Local users
  • Active users
  • Balance between too many and too few matches
  • Low latency prioritized over real-time recommendations
• Geo Sharding for Locality
  • Using latitude and longitude for database sharding
  • Relies on services like Google S2 for shard management
  • Recommendation Flow:
    • Client requests recommendation list
    • Geo Sharding Mapper retrieves local shards
    • Filters profiles based on preferences and activity
    • Returns a limited number of profiles (e.g., 50)
• Recommendation Service Tasks:
  • Filter profiles based on preferences & active status
  • Ensure profiles don't exceed a set limit
  • Balance matches across users

Considerations for Optimization

• Scalability:
  • Load balancers
  • Memory caches (e.g., Redis)
• Storage & Processing Efficiency:
  • Image resizing/encoding
  • NoSQL databases for flexibility and scale

Conclusion & Tips for Interviewees

• Key Interview Strategy:
  • Start broad, cover core functionalities
  • Dive deep into individual components
  • Balance deep dives with broad coverage
• Interviewer Role:
  • Guide the discussion
  • Keep interviewees on track when needed

Final Thoughts

• Focus on a balanced approach between low latency and quality recommendations.
• Prioritize user experience (e.g., low waiting time for recommendations).
• Be prepared to discuss core functionalities distinctively and integratively.