Loading...
Meta
S17
HardPremiumDesign a Trust & Safety / Content Moderation Agent at Scale
Design a trust and safety system that uses LLMs to moderate user-generated content on a platform with millions of posts per day.
ClassificationSafetyScaleCost Optimization
Key Requirements
- Catch harmful content (hate speech, spam, CSAM) with high recall
- Keep false positives low to avoid frustrating legitimate users
- Real-time moderation at millions-of-posts-per-day scale
- Appeals workflow for wrongly flagged content
- Adversarial robustness against users trying to evade detection
Interviewer Follow-ups
- Q1How do you handle content that is borderline or context-dependent?
- Q2How do you keep up with new types of abuse?
- Q3How do you handle appeals efficiently at scale?