Product design question

How would you build a system that detects adult content on the twitter platform?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

System design for content moderation: balancing detection accuracy, latency, user experience, and false positive cost at platform scale.

Clarify scope: images and video uploaded to tweets, in DMs, or both, and confirm with the interviewer.
Identify the goal: minimize adult content exposure to minors and unwilling viewers without over blocking legitimate content.
Design the pipeline: automated classifier at upload time, followed by human review for borderline cases.
Decide on enforcement: auto blur with a warning screen for medium confidence, removal for high confidence.
Define success with recall on known adult content, false positive rate, and review queue latency.
Note the tradeoff: aggressive filtering risks flagging art or health content, so appeals must be fast.

Separates automated detection from human review rather than proposing one single black box system.
Names a specific enforcement tier, blur plus warning for medium confidence, instead of binary allow or block.
Sets a guardrail, false positive rate on legitimate content like art or medical images.
Assumes an illustrative target, such as 95 percent recall with sub one percent false positive rate.

Describes only the machine learning model with no product or policy layer.
Ignores appeals and false positives, which are the real world cost of this system.

How would you handle borderline artistic or medical content?
How would you keep the human review team from burning out on this content?
How would you measure the model's performance across different regions and cultures?