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

The ALeRCE text-to-SQL system leverages large language models (LLMs) to facilitate querying the ALeRCE astronomical database through natural language, translating it into executable SQL queries. The system incorporates a four-module framework—schema linking, query classification, prompt decomposition, and self-correction—and was evaluated using a dataset of 110 natural language/SQL pairs. Notably, Claude Opus 4.6 achieved a perfect-match performance of 0.97 and 0.94 for row and column identifiers on simple queries, with performance declining for more complex queries, highlighting the importance of model selection and architectural enhancements in improving text-to-SQL capabilities for practical applications in data querying.