Biome is an outdoor-focused social platform built for Gen Z, where users share live photo moments with friends and communities. As the platform expanded to thousands of users with 200 to 300 concurrent users during peak periods, manual content moderation became increasingly difficult to scale. Biome needed an AI content moderation system to automate image categorization, protect younger users from harmful content, and accelerate content visibility while maintaining community standards.
Problem: Manual moderation slows AI content moderation at scale
Biome manually reviewed every uploaded image before publishing it to the platform. Moderators classified each post into one of five permission categories: Circle, Community, Private, Hide, or Delete.
As content volume increased, this workflow created operational bottlenecks, delayed content visibility, and increased moderation effort. Since the platform serves users aged 13 to 18, moderators also needed to carefully detect explicit and harmful images before they appeared in feeds. The manual process limited scalability and made consistent moderation increasingly difficult.
Solution: AI content moderation for automated image classification
GoML proposed an AI content moderation system using its Agentic AI blueprint to automate visual content analysis while integrating directly with Biome's AWS infrastructure.
The solution continuously detects newly uploaded posts, retrieves images from Amazon S3, analyzes visual content using foundation models on Amazon Bedrock, and automatically assigns the appropriate permission category. The AI content moderation engine operates independently from Biome's backend, updates moderation decisions asynchronously, and continuously improves using historical moderation data.
AI content moderation for image categorization
Key capabilities:
• Continuously monitor new uploads from Amazon RDS MySQL
• Retrieve images securely from Amazon S3
• Analyze visual content using Amazon Bedrock foundation models
• Classify images into Circle, Community, Private, Hide, or Delete
• Automatically identify explicit and harmful visual content
• Process images asynchronously without affecting application performance
• Update moderation decisions directly in the database
AI-powered content safety detection
Key capabilities:
• Detect explicit and harmful visual content
• Protect younger users from unsafe images
• Prevent inappropriate content from appearing in community feeds
• Apply AI content moderation policies consistently
• Improve moderation quality using moderator feedback
• Validate predictions against historical moderation decisions
Independent AI content moderation workflow
Key capabilities:
• Python-based moderation service
• Autonomous polling every one to two minutes
• Parallel processing independent of the backend
• Secure database read and write operations
• Asynchronous image processing pipeline
• No direct API dependencies with the application
Continuous model improvement
Key capabilities:
• Train models using historical moderation datasets
• Benchmark predictions against human moderation decisions
• Improve AI content moderation accuracy through feedback loops
• Validate categorization quality before deployment
• Support ongoing model retraining
• Reduce manual moderation effort over time
Impact
- 80%+ image categorization accuracy against historical moderation data
- 90%+ explicit content detection accuracy
- 99%+ successful image retrieval from Amazon S3
- Automated moderation every one to two minutes
- Reduced manual moderation workload
- Faster content visibility while maintaining platform safety
About
Before Gen AI and after Gen AI
"With AI content moderation embedded into the content pipeline, Biome transformed manual image review into an intelligent, scalable moderation workflow that accelerates content visibility while protecting community safety."
Prashanna Rao, Head of Engineering, GoML
Key takeaways for AI content moderation platforms
Common challenges
- Manual moderation struggled to keep up with growing image uploads.
- Harmful content detection relied heavily on human reviewers.
- Moderation delays slowed content publishing.
- Higher content volumes increased moderation costs.
- Maintaining platform safety required constant manual oversight.
Practical guidance
- Automate image classification with AI while retaining human review for exceptions.
- Train models using historical moderation decisions.
- Process uploads asynchronously for faster moderation.
- Integrate AI moderation into the existing cloud infrastructure.
- Continuously improve models with moderator feedback.
Ready to modernize content moderation?
Partner with GoML to build scalable AI content moderation systems with faster delivery through AI Matic.
