Retail & Media
Top content discovery platform - automated moderation
Millions of daily content submissions across multiple languages and formats, with manual moderation unable to keep pace
The challenge
A top content discovery platform struggled to moderate millions of daily submissions across multiple languages and content types. Manual moderation couldn't keep pace, and maintaining consistent policy-aligned tagging was increasingly difficult and cost-prohibitive at the platform's scale.
What we built
We built a closed-loop ML and deep learning pipeline that predicts moderator decisions with high accuracy. The system uses multi-modal modeling over text and images, implemented in TensorFlow with custom embeddings, enabling automatic review at massive scale while continuously improving from moderator feedback on edge cases.
What changed
Of content automatically moderated
Items processed across multiple languages
Moderation cost and staffing requirements
Built with
TensorFlow · Deep Learning · Multi-modal AI · NLP · Computer Vision · Closed-loop Learning