Moderation Filter

Overview:

The system evaluates uploaded images for unsafe content using Rekognition moderation labels and enforces a blocking policy when necessary.

Prerequisites:

• Existing Lambda pipeline
• Moderation permissions
• Defined payload contract with moderation fields
• Agreement on the moderation policy behavior

Implementation/Steps:

1. The system invokes detect_moderation_labels alongside label detection.
2. The system normalizes moderation labels and sorts them by confidence.
3. The system sets a flagged state when moderation thresholds are met.
4. The system blocks flagged images by returning a structured response.
5. The system excludes normal tags when content is flagged.

Key Takeaways

• Moderation runs by default with optional configuration.
• Flagging follows a single consistent rule.
• Unsafe content is blocked at the pipeline level.
• Moderation output remains separate from normal tags.