Mantel is a social platform for collectors of cards, comics, coins, and other collectibles. It connects users through content sharing, discussions, and discovery of rare items. The platform currently uses a chronological feed and aims to improve user engagement through a personalized experience. At present, Mantel has built a strong data foundation and is preparing for a future content recommendation system.
Problem: lack of personalization limits content discovery
Mantel relied on a chronological feed that displayed content purely based on time rather than relevance, which limited the overall user experience. Users often struggled to discover posts, collections, and discussions that matched their interests, especially as the volume of content grew. The platform lacked personalization based on user behavior or preferences, resulting in low engagement, particularly for new users who were not immediately shown relevant content. Without intelligent ranking, valuable and niche content remained hard to find, and the system could not effectively adapt to individual user needs. This absence of a content recommendation system made it difficult for Mantel to scale content discovery and maintain strong user retention.
Solution: data foundation for Gen AI powered content recommendation system
GoML built a scalable data pipeline and structured knowledge base, using AI Content Generation Accelerator, to prepare Mantel for a content recommendation system, instead of directly replacing the chronological feed with an intelligent, personalized feed based on user preferences and engagement signals. This data foundation is designed to support future machine learning models, ranking systems, and user behavior analysis for delivering relevant content
Content recommendation system and feed ranking
- Personalized feed based on user interests
- Dynamic ranking of content based on relevance
- Real time adaptation to user behavior
- Better discovery of niche and rare collectibles
- Improved first time user experience
User preference modeling
The system is designed to capture and use multiple signals in future implementations:
- User onboarding preferences
- Content interactions such as likes, views, and comments
- Behavior patterns across sessions
- Interest based segmentation
Content ranking and relevance scoring
The system enables future content ranking using:
- Machine learning based ranking algorithms
- Relevance scoring based on user signals
- Candidate generation for content selection
- Real time ranking updates
Data foundation and pipeline support
The recommendation system is supported by a strong data pipeline:
- Structured trading card knowledge base
- Hierarchical data storage across sport, year, edition, set, and cards
- Automated data ingestion and validation
- Reliable data extraction using a web scraping pipeline
Interactive user experience
The platform moves from static feed to intelligent interaction:
- Feed updates dynamically based on user activity
- Content relevance improves over time
- Seamless discovery of new and relevant posts
- Consistent experience across sessions
Infrastructure and deployment
The system uses a scalable cloud stack:
- AWS infrastructure with ECS, Lambda, and EventBridge
- PostgreSQL database for structured data
- CloudWatch for monitoring and logs
- Containerized deployment using Docker and ECR
Quality assurance
Validation focused on performance and accuracy:
- Testing ranking accuracy and relevance
- Validating personalization signals
- Ensuring stable feed updates
- End to end testing with real user scenarios
Impacts
- 60-70% improvement in data readiness for personalization
- 50-65% improvement in data quality and structure
- 40-50% faster development of recommendation features
- 2x-3x scalability for future personalized experiences
About
Before Gen AI and after Gen AI
“With a content recommendation system, Mantel is now prepared to transform its feed into a personalized experience that connects collectors with the most relevant content.”
Prashanna Rao, Head of Engineering, GoML
Key takeaways for social platforms
Common challenges
- Chronological feeds reduce relevance
- Lack of personalization impacts engagement
- Difficulty scaling content discovery
Practical guidance
- Adopt a content recommendation system for personalized feeds
- Use user behavior and preferences for ranking
- Implement real time relevance scoring
- Start with high impact use cases like feed ranking
Ready to build content recommendation systems
Partner with GoML to build scalable AI powered content recommendation systems using Gen AI with AI Matic.
