Weaviate Researchers Introduce Function Calling for LLMs: Eliminating SQL Dependency to Improve Database Querying Accuracy and Efficiency

Databases are essential for storing and accessing organized data, but querying them can be complex, especially with SQL. Large language models (LLMs) can help automate this process, but they often struggle with SQL syntax. To improve this, a new API approach is being developed to help LLMs interact with databases more effectively. This aims to enhance the accuracy of queries. Current text-to-SQL solutions face challenges like different SQL dialects, complex queries, and the need for accurate database targeting. A new benchmark called DBGorilla has been created to address these issues. It allows LLMs to query databases without SQL by using defined API functions for searching and filtering. The DBGorilla dataset includes 315 queries across five database schemas and evaluates performance based on accuracy and collection routing. Tests showed that top models like Claude 3.5 Sonnet achieved over 74% accuracy in matching queries and high routing accuracy. The study indicates that function calling is a strong alternative to traditional text-to-SQL methods. Companies looking to implement AI should identify automation opportunities, set measurable goals, choose suitable solutions, and start with pilot projects. For further assistance or insights on AI solutions, contact us at hello@itinai.com.