You are designing the ad ranking system for a social feed with 200 million daily active users. Every feed request must select which ads to show among roughly 1 million eligible campaigns, and it must do so in under 50 milliseconds because the ad request sits inside the feed request's latency budget.
Advertisers bid on conversions (an app install, a purchase), not clicks. You charge per impression on a second-price basis, which means the system must convert a conversion bid into an expected value per impression, and that conversion happens hours or days after the impression, if it happens at all.
Revenue is the obvious objective, but showing too many bad ads degrades the feed and users leave. Leadership wants both protected.
Build the architecture on a canvas: place the components, configure them, connect them into a data flow, and write a short reason for each one. The AI reviewer grades your design against a rubric written specifically for this problem.
Your CTR model ranks perfectly but predicts probabilities 2x too high. What breaks?
An impression today might convert in 3 days. How do you build tomorrow's training set without teaching the model that recent traffic never converts?
A campaign spends its entire daily budget in the first hour. What is wrong, and how does your pacing fix it?
Minimum 8 components · needs a wide desktop screen