Shaped

Shaped is a web and API platform for businesses building search, recommendations, feed ranking, and agent retrieval that need real-time personalization without maintaining complex retrieval infrastructure.

Key capabilities include:

• Real-time retrieval and ranking with sub-50ms query latency and a REST API for serving personalized results.
• ShapedQL, a SQL-like query language that compiles into multi-stage ranking pipelines, including retrieve, filter, score, and reorder steps for relevance, business rules, and diversity.
• Data connectivity and transforms with 30+ connectors (including Snowflake, Postgres, Shopify, Kafka), plus SQL transforms/materialized views and AI enrichment transforms (LLM-based embedding generation and structured extraction).
• Modeling for unstructured and multi-modal data, using transformer and LLM-based techniques to rank text, images, and video, with continuous learning from behavioral signals.

Notable for combining data ingestion, enrichment, model training, and a declarative ranking layer in one system, plus SOC 2 Type II posture and SLA targets, positioning it as an alternative to piecing together Elasticsearch/Vector DBs, feature pipelines, and custom ranking services.