Structured and Unstructured Query Language (SUQL)
Conversational Search over Structured and Unstructured Data with LLMs
It augments SQL with several important free text primitives for a precise, succinct, and expressive representation. It can be used to build chatbots for relational data sources that contain both structured and unstructured information. Similar to how text-to-SQL has seen great success, SUQL can be used as the semantic parsing target language for hybrid databases, for instance, for:
An example restaurant relational database
Several important features:
-
SUQL seamlessly integrates retrieval models, LLMs, and traditional SQL to deliver a clean, effective interface for hybrid data access;
- It utilizes techniques inherent to each component: retrieval model and LM for unstructured data and relational SQL for structured data;
-
Index of free text fields built with faiss, natively supporting all your favorite dense vector processing methods, e.g. product quantizer, HNSW, etc.;
- A series of important optimizations to minimize expensive LLM calls;
- Scalability to large databases with PostgreSQL;
- Support for general SQLs, e.g. JOINs, GROUP BYs.
Tags: language storage ai
Last modified 16 December 2024