System Design Interview
Production-grade TikTok architecture with creator upload flows, transcoding, low-latency playback, For You ranking, creator analytics, moderation/copyright enforcement, ads/live commerce, and global platform resilience.
Start with the fundamentals: upload a video, store metadata, and play it back through a simple API.
What was missing: Transcoding, CDN, recommendations, counters, and moderation.
Why that's risky: Raw origin playback cannot handle adaptive playback or global spikes.
What gets added: Edge, video API, raw media storage, and canonical metadata store.
Trade-offs: Simple but not globally performant.
Add the actual video platform backbone: async media processing, ABR renditions, manifests, and CDN delivery.
What was missing: Adaptive bitrate delivery and asynchronous processing.
Why that's risky: Without a quality ladder and CDN, playback suffers badly on mobile networks and global traffic spikes.
What gets added: Upload session service, processing queue, media workers, rendition store, CDN, and playback metadata.
Trade-offs: More infra and media-pipeline complexity.
Introduce the core TikTok differentiator: a personalized feed built from retrieval, ranking, and engagement signals.
What was missing: Personalization, candidate generation, and realtime learning from engagement.
Why that's risky: A generic feed destroys retention and makes TikTok feel like a simple video host, not a recommendation product.
What gets added: Feed API, feed cache, retrieval service, online ranker, engagement stream, online feature store, and reco store.
Trade-offs: ML complexity, feature freshness issues, and more infra cost.
Add the systems that turn viewers into creators and viral feedback loops into product features.
What was missing: Engagement aggregation, creator feedback loops, and viral trend systems.
Why that's risky: Without strong analytics and trend systems, the platform can’t amplify creators or learn from user behavior effectively.
What gets added: Engagement API, creator analytics API, engagement stream, counter/trend/analytics workers, and dedicated stores for counters, trends, and creator insights.
Trade-offs: More async consistency and heavier offline/nearline analytics cost.
Add the heavy safety systems a serious short-video platform must run continuously.
What was missing: Trust, safety, and copyright systems.
Why that's risky: Short-video platforms are abuse magnets; without strong moderation and copyright workflows, quality and legal risk spiral quickly.
What gets added: Secure edge, policy API, risk store, moderation queue, safety/copyright/abuse workers, enforcement store, and human review console.
Trade-offs: Moderation cost, model drift, false positives, and operational overhead.
Finish with the full platform view: monetization, live systems, analytics, multi-region control, and safe operations.
What was missing: Monetization, live commerce, experimentation, and global operational control.
Why that's risky: A TikTok-scale product is also an ads, creator-economy, and global ops platform—not just a feed app.
What gets added: Global routing, multi-region platforms, event backbone, feature flags/config, analytics warehouse, ad serving, live-commerce service, model training, observability, and backups.
Trade-offs: Significant platform complexity, cost, and consistency tradeoffs.
Use ABR with HLS/DASH, small segments, quick-start low-bitrate renditions, aggressive CDN edge caching, and client prefetch of the next likely video.
Run layered moderation: reputation checks at upload time, async policy classifiers, copyright/audio fingerprint matching, and human review/takedown workflows.
Use CDN origin shielding, autoscale feed and metadata services, coalesce hot requests, and prioritize prefetch/caching of manifests and hot segments for the viral object.
Combine trending content, regional priors, demographic-safe bootstrapping, creator diversity, and simple onboarding signals until enough watch/skip/dwell data exists for personalization.
Retrieval finds a broad candidate pool quickly using embeddings, trending pools, and heuristics. Ranking then applies more expensive scoring on a smaller set using online features, quality signals, and safety rules. Separating them keeps latency tractable.
Stream engagement events asynchronously, aggregate them nearline/offline, and serve creators from a dedicated analytics store rather than reading directly from operational counters or the watch path.
Insert sponsored candidates in a separate ad-serving step with pacing, auction, and policy constraints. Apply strict latency budgets, relevance thresholds, and frequency caps; fall back gracefully if ad systems are unavailable.
Keep playback and manifest delivery independent from ranking and moderation pipelines. Recommendation can degrade to cached or trending pools, and moderation decisions can apply asynchronously with emergency blocks for high-risk content.
You're in the middle of an interview session. Leaving now will end your current attempt.
Explore concept overviews, real-system examples, key tradeoffs, and interview talking points for each roadmap section.
You've conquered this phase. These are the skills you now own:
