Querying an astronomical database using large language models: the ALeRCE text-to-SQL system

2026-06-16 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Astrophysical Data Systems · Depth: Advanced, quick

Summary

The ALeRCE text-to-SQL system, developed using large language models (LLMs) and in-context learning, enables natural language querying of the ALeRCE astronomical database, a community broker for the Zwicky Transient Facility and Vera C. Rubin Observatory. To facilitate its development and evaluation, a dataset of 110 natural language/SQL pairs was constructed. The system employs a four-module step-by-step generation framework: schema linking, query classification, prompt decomposition, and self-correction, which consistently outperforms a direct-inference baseline. Evaluation of thirteen LLMs showed that the self-correction module significantly reduces execution errors. For Claude Opus 4.6, perfect-match (PM) performance on row (column) identifiers reached 0.97 (0.94) for simple queries, decreasing to 0.44 (0.72) for medium queries and 0.59 (0.49) for hard queries. Top-performing LLMs included Claude Opus 4.6, Gemini 2.5 Pro, Gemini 3 Flash, and GPT-5.2-Codex.

Key takeaway

For AI Engineers implementing natural language interfaces for complex databases like astronomical archives, you should prioritize a multi-module, step-by-step generation framework over direct inference. Incorporating a self-correction module is critical to significantly reduce execution errors in generated SQL queries. When selecting an LLM, consider top performers like Claude Opus 4.6, Gemini 2.5 Pro, Gemini 3 Flash, or GPT-5.2-Codex, especially for handling queries of varying complexity.

Key insights

The ALeRCE text-to-SQL system uses LLMs and a step-by-step framework to query astronomical databases via natural language.

Principles

• Step-by-step generation outperforms direct inference.
• Self-correction modules reduce execution errors.
• Query complexity impacts text-to-SQL performance.

Method

A four-module framework: schema linking, query classification, prompt decomposition, and self-correction, generates executable SQL from natural language queries.

In practice

• Query astronomical databases with natural language.
• Evaluate LLMs for text-to-SQL tasks.
• Improve query accuracy via self-correction.

Topics

• Text-to-SQL
• Large Language Models
• Astronomical Databases
• ALeRCE System
• Prompt Engineering
• Self-correction

Best for: NLP Engineer, AI Scientist, AI Engineer, Machine Learning Engineer, Research Scientist

Related on AIssential

• Your Database Knows Everything. Too Bad It Only Speaks SQL. — An open-source Agentic AI module has been developed to enable users to query databases like Redshift and SQLite using plain English, eliminating the need for manual SQL writing for common tasks.
• Teaching Your SQL Generator to Fix Its Own Mistakes — LLMs can self-correct runtime SQL errors by processing database feedback in a retry loop.
• Semantic adapters in text-to-SQL for low-resource languages: the importance of semantic information — Injecting semantic structural knowledge into LLMs significantly boosts Text-to-SQL performance for low-resource languages.
• DecoSearch: Complexity-Aware Routing and Plan-Level Repair for Text-to-SQL — DecoSearch improves Text-to-SQL accuracy for complex queries via complexity-aware routing and plan-level repair.
• TAHOE: Text-to-SQL with Automated Hint Optimization from Experience — TAHOE optimizes Text-to-SQL LLM prompts by dynamically learning and applying error-driven hints from development and deployment experiences.
• Multi-Agent Architecture with RAG and Dynamic Context Windows for Text-to-SQL Optimization — A multi-agent Text-to-SQL system uses RAG and dynamic context to reduce token consumption.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.