LLM Content Moderation System

Problem Statement

A social media platform needs to scale content moderation from 10K to 1M daily posts. Current human review can't keep up, and rule-based systems miss nuanced violations like coded language and contextual harassment.

Scope

• •Map the current moderation workflow and identify AI insertion points
• •Design the LLM-based moderation pipeline with human-in-the-loop
• •Define category taxonomy and severity levels
• •Create evaluation framework with precision/recall targets
• •Write the content policy that the AI system will enforce

Evaluation Rubric

• •Understanding of safety/trust tradeoffs
• •Pipeline architecture quality
• •Evaluation framework thoroughness
• •Policy clarity and edge case handling
• •Scalability and operational considerations