AI Moderation via REACH - Content Moderation

AI Moderation via REACH

REACH (Real-time Enrichment, Analysis, and Classification Hub) provides AI-powered content moderation using LLMs and computer vision models. Content is analyzed against configurable policies that define rules for text, visual, and audio content.

Ordering a Moderation Task

curl -X POST "$KALTURA_SERVICE_URL/service/entryVendorTask/action/add" \
  -d "ks=$KALTURA_KS" \
  -d "format=1" \
  -d "entryVendorTask[objectType]=KalturaEntryVendorTask" \
  -d "entryVendorTask[entryId]=$KALTURA_ENTRY_ID" \
  -d "entryVendorTask[catalogItemId]=$MODERATION_CATALOG_ITEM_ID" \
  -d "entryVendorTask[taskJobData][objectType]=KalturaEntryVendorTaskModerationJobData"

Moderation catalog items have serviceFeature=15. List available items:

curl -X POST "$KALTURA_SERVICE_URL/service/vendorCatalogItem/action/list" \
  -d "ks=$KALTURA_KS" \
  -d "format=1" \
  -d "filter[serviceFeatureEqual]=15"

Reading Moderation Results

Poll the task until status=2 (READY):

curl -X POST "$KALTURA_SERVICE_URL/service/entryVendorTask/action/get" \
  -d "ks=$KALTURA_KS" \
  -d "format=1" \
  -d "id=$VENDOR_TASK_ID"

The moderation output is in the moderationOutputJson field of the task's taskJobData:

{
  "moderationOutputJson": {
    "violations": [
      {
        "rule": "profanity",
        "severity": "high",
        "timestamp": 45.2,
        "description": "Explicit language detected in transcript"
      }
    ],
    "summary": "Content flagged for profanity at 0:45",
    "overallScore": 0.85,
    "policyId": "default_moderation_v2"
  }
}

Task status values (KalturaEntryVendorTaskStatus):

Value	Name	Description
1	PENDING	Task queued for processing
2	READY	Analysis complete — results available
3	PROCESSING	Currently being analyzed
5	ERROR	Task failed — check errDescription
7	ABORTED	Cancelled before completion

Category Auto-Action

When a moderation task completes, the system can automatically activate or reject the entry within a specific category based on the moderation result:

• Pass → categoryEntry.activate — entry becomes visible in that category
• Fail → categoryEntry.reject — entry is hidden in that category

This is configured at the catalog item level and applies per-category, not per-entry. The entry's global moderationStatus is not changed by this auto-action — only the categoryEntry status is affected.