We use cookies for site analytics. Accept to help us understand how the site is used. See our Privacy Policy for details.
The sub-100ms auction, candidate ad retrieval, pCTR/pCVR prediction, second- vs first-price auctions, budget pacing, frequency capping, attribution, fraud, and the money-correctness problem of never overspending a budget.
Design the platform that decides, in the ~80 milliseconds while a web page or app screen is loading, which ad to show a specific user - and what it costs. Millions of ad requests per second, millions of active campaigns each with a budget and targeting rules, an auction to pick a winner, a price to charge, and a hard promise to every advertiser: we will spend your budget smoothly and never a cent over.
This is system design with a money-correctness twist. It combines the hardest parts of several problems: a low-latency serving path (like search/recommendations), an ML ranking problem (predict click and conversion probability), a market-design problem (the auction), and a financial-correctness problem (budgets, pacing, billing) where being wrong costs real money in both directions - overspend and you eat the loss, underspend and the advertiser churns. Strong candidates keep the auction inside the latency budget while never breaking the budget-correctness invariant.
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 · 5 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.
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.
Online vs batch inference, GPU utilization tricks, autoscaling for spiky load, A/B testing models, and the feature store that decouples training from serving.
Idempotency keys, double-spend prevention, the ledger model, and why eventual consistency is wrong for balances. The interview where ambiguity costs you money.
Inverted indexes, BM25 ranking, prefix tries, and the p99 < 100ms latency budget that drives every architectural choice.
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 →