📝 Problem Description
Design a short video platform like TikTok. Users can record/upload videos, scroll through personalized "For You" feed, like, comment, and share. Focus on video processing, recommendation algorithm, and low-latency delivery.
👤 Use Cases
1.
Creator wants to uploads video so that video processed and published
2.
User wants to scrolls For You page so that sees personalized video feed
3.
User wants to likes or comments so that engagement recorded
4.
Algorithm wants to analyzes engagement so that adjusts recommendations
✅ Functional Requirements
- •Record and upload short videos (15s-3min)
- •Video editing (filters, music, effects)
- •Personalized "For You" feed
- •Following feed
- •Like, comment, share, save
- •Duets and stitches
- •Trending hashtags and sounds
⚡ Non-Functional Requirements
- •Video upload complete < 30 seconds
- •Instant video playback (< 500ms)
- •Support 1B monthly users
- •100M concurrent users
- •99.9% availability
⚠️ Constraints & Assumptions
- •Videos must be transcoded to multiple qualities
- •ML-heavy recommendation engine
- •Global content moderation required
📊 Capacity Estimation
👥 Users
1B MAU, 100M DAU
💾 Storage
500PB (videos)
⚡ QPS
Uploads: 100K/sec, Views: 10M/sec
🌐 Bandwidth
100Pbps peak
📐 Assumptions
- • 1B monthly active users
- • 100M daily active users
- • Average 1 hour watch time per day
- • 5M videos uploaded per day
💡 Key Concepts
CRITICAL
ML Recommendation
Multi-stage ranking: candidate generation → scoring → diversity → serving.
CRITICAL
Video Transcoding
Transcode to multiple bitrates (240p-1080p) for adaptive streaming.
HIGH
Real-time Feedback
Watch time, likes instantly feed back to ranking model.
HIGH
Content Moderation
ML + human review pipeline before videos go live.
💡 Interview Tips
- 💡Start with the recommendation algorithm
- 💡Discuss the For You page personalization
- 💡Emphasize the content moderation pipeline
- 💡Be prepared to discuss video processing
- 💡Know the difference from YouTube (short-form)
- 💡Understand the creator economy features