We use cookies for site analytics. Accept to help us understand how the site is used. See our Privacy Policy for details.
The two-stage candidate-generation-then-ranking architecture, embeddings + ANN retrieval, the batch/real-time feature split, feedback loops, cold start, and why offline NDCG lies until the online A/B disagrees.
Design the recommendation system that fills a personalized feed - the YouTube home page, the Netflix row grid, the TikTok For You feed. Hundreds of millions of users, a catalog of billions of items (or short videos uploaded by the second), and a hard requirement that the feed feels fresh, relevant, and responsive to what the user just did.
This is the flagship "big-tech ML system design" question. Weak candidates describe "a model that predicts what the user will like" as if it were one model call over a billion items - which is computationally impossible at request time. Strong candidates reach immediately for the two-stage funnel: cheap candidate generation that narrows billions to hundreds, then an expensive ranking model that orders those hundreds. Excellent candidates own the feedback loop - the recommender trains on data the recommender itself generated - and can articulate why that creates filter bubbles, popularity runaway, and offline/online metric divergence.
Asking these before diving into a solution is the difference between a "hire" and a "no signal" rating. Pick the questions whose answers would change your design.
Capacity estimation · architecture with all 9 components explained · 6 deep dives · trade-off analysis · 8 common follow-up questions
Get full access to gitGood.dev with a subscription.
Try everything free for 10 days. Cancel anytime.
10-day trial, then $5/mo - or $40/yr ($3.33/mo, save 33%). Cancel anytime.
Online vs batch inference, GPU utilization tricks, autoscaling for spiky load, A/B testing models, and the feature store that decouples training from serving.
Inverted indexes, BM25 ranking, prefix tries, and the p99 < 100ms latency budget that drives every architectural choice.
The classic write-vs-read amplification trade-off. Push, pull, or hybrid fanout - and how to handle the celebrity user with 100M followers.
Batch vs streaming, lambda vs kappa, the warehouse-vs-lakehouse decision, and dimension modeling that survives schema drift.
Reading is the floor. The interview signal is in walking through this live with someone probing follow-ups. Use the AI mock interview to practice talking through requirements, architecture, and trade-offs out loud.
Start an AI mock interview →