Ads bidding and ranking

Design the ML system behind an ads-targeting and real-time-bidding service that, for each ad opportunity, selects and ranks candidate ads and computes a bid within a 50ms p99 budget at 100k requests/sec. It must predict click and conversion probability, respect per-advertiser budgets and pacing, and run a second-price-style auction. Cover candidate selection, the prediction models, how bids and budgets are managed, and the calibration requirement.