AI Content Moderation
What It Does
Real-time text, image, and video moderation for user-generated-content platforms. It catches obscene content automatically and flags grey-zone cases for admin review — context-aware, rather than relying on keyword bans alone.
How It Works
01 — Keyword Filter: a regex and blocklist pass that catches 30–40% of cases with no AI cost.
02 — Text Moderator (OpenAI omni-moderation-latest): scores across 11 violation categories.
03 — Image Moderator (AWS Rekognition DetectModerationLabels): applies hierarchical confidence labels.
04 — Video Moderator (AWS Rekognition StartContentModeration): moderates frame by frame.
05 — Decision Engine: combines the modality results into ALLOW, BLOCK, or REVIEW.
06 — Grey-Zone Router: sends ambiguous cases to an admin queue, and feedback tunes thresholds.
07 — Audit Logger (PostgreSQL append-only): records every decision with metadata.
Human-in-the-loop: admins review grey-zone cases, tune thresholds, and work from trust-and-safety dashboards.
Screen every post across text, image, and video in one workflow, escalating only the calls that need a person.
- Text, image, and video in one flow
- Hybrid cost/accuracy methodology
- Context-aware, not blocklist-only
- Decision logs with confidence scores
- SR&ED-eligible development
- Real-time API or batch
Why AI Content Moderation
Thousands per hour
Move from roughly 50 posts per hour per moderator to thousands per hour.
~70% lower cost
Automating the obvious cases cuts moderation cost by around 70% versus a full manual team.
Near-instant decisions
Sub-second for text, about 3 seconds for images, and roughly 10 seconds per minute of video.
Judgement where it matters
Grey-zone cases route to admins with confidence scores, so context-sensitive calls stay human.