Key features

info

This is a preview of the new Shaped docs. Found an issue or have feedback? Let us know!

Shaped is a platform for building, deploying, and scaling recommendation and search systems. It provides AI models, data infrastructure, and APIs for discovery use cases.

Connector library

Shaped supports over 20 connectors to train your engines. Connect Shaped to your data sources using a declarative interface. Automated data pipelines handle ingestion and keep your recommendation system updated with the latest data.

Real-time processing

Shaped ingests and processes interaction data through streaming pipelines, enabling dynamic reranking based on user preferences, trending content, and other real-time signals. See the guide to real-time connectors for more information.

Model library

Shaped's model library includes:

• Fine-Tuned LLMs: Fine-tuned large language models for content understanding and personalized ranking.
• Neural Ranking Models: Neural networks optimized for ranking tasks.
• Multi-Stage Systems: Shaped combines multiple models and techniques—content understanding, retrieval, scoring, and ordering—into multi-stage recommendation systems.

Customization

Configure Shaped for your use case by combining and configuring ranking and retrieval components. For example, see: metadata filtering, result diversity, and exploration factors.

Test different model configurations and ranking strategies to optimize for your objectives.

Multi-Objective Optimization

Most recommendation and search use cases optimize for multiple objectives. In addition to primary engagement objectives (e.g., clicks) or conversion objectives (e.g., purchases), you may also optimize for diversity, cart size, or different stakeholder groups (e.g., buyers and sellers in a marketplace).

Shaped frames multi-objective optimization as:

1. Top-level metrics (e.g., accuracy, diversity, novelty, serendipity)
2. User, item, or event slices (e.g., creators, consumers, sellers, buyers, new items, old items, purchases)

Shaped optimizes across multiple top-level metrics and data slices, ensuring balanced trade-offs across user cohorts, item taxonomies, and event types during training.

Analytics and Metrics

Shaped provides:

• In-Session Analytics: Visualize user behavior and recommendation patterns in real time.
• Performance Metrics: Track metrics like recall, precision, and coverage. Breakdowns by data segment (e.g., new users, popular items) identify strengths and weaknesses. See the guide on evaluating your model.

Infrastructure

Shaped's infrastructure provides:

• Scalability: Handle millions of users and items with a scalable architecture.
• Security: Enterprise-grade security measures, including GDPR and SOC2 compliance. For more details, see the security page.

Shaped provides recommendation and search infrastructure, models, and APIs for building discovery systems.