Build content moderation API with transformers

Task

Implement AI-powered content moderation service

Requirements

• Use HuggingFace transformers for text classification
• Multi-label classification (toxic, spam, harassment, etc.)
• Support multiple languages (EN, ES, FR, DE)
• Real-time inference with <200ms latency
• Explainability via attention weights

Models to Evaluate

• RoBERTa-large for toxicity
• DistilBERT for speed
• Multilingual BERT for i18n

API Design

• POST /api/v1/moderate - Analyze text content
• POST /api/v1/moderate/batch - Batch processing
• Response includes labels + confidence scores

Integration

• Android and iOS apps will call this API
• Must handle high throughput (1000s req/min)

Estimated effort

1.5-2 weeks