Shaped - A SQL engine for multi-stage ranking and RAG

What it does

ShapedQL is a SQL engine designed for multi-stage ranking and retrieval-augmented generation (RAG). It allows users to write queries that combine lexical search, semantic search, hybrid search, image search, and personalized reranking. The playground demonstrates ShapedQL against a live demo model using the Movielens dataset, which includes movie titles, descriptions, posters, user interactions, and personnel data. Users can test queries with various embeddings (title, description, poster, collaborative, personnel) and scoring models like a click-through rate predictor.

Who it is for

ShapedQL targets developers and data scientists building search, recommendation, or RAG systems. It is particularly useful for those who need to combine multiple ranking stages (e.g., semantic search followed by personalized reranking) in a single query language. The playground is aimed at users evaluating ShapedQL for their own datasets and use cases.

Why it matters

Traditional SQL engines are not optimized for ranking and retrieval tasks, especially when combining multiple signals (text, image, user behavior). ShapedQL provides a unified query language that simplifies the development of complex ranking pipelines. By supporting multi-stage ranking and RAG, it enables more relevant and personalized results without requiring custom code for each stage.

Launch signal

ShapedQL was introduced via a Show HN post on Hacker News, indicating an early-stage launch. The playground is publicly accessible, and documentation is available at docs.shaped.ai. The product appears to be in beta or early access, with a registration link for loading custom datasets.

Brand and naming

The name "Shaped" suggests shaping or molding results to user needs, aligning with its focus on ranking and personalization. The tagline "A SQL engine for multi-stage ranking and RAG" clearly communicates its core value proposition. The brand positions itself as a developer tool that extends SQL capabilities for modern AI-driven search and recommendation tasks.