How Dataherald Makes Natural Language to SQL Easy

Text-to-SQL is an important use case for developers working with relational data. However, modern Language Learning Models (LLMs) have been found to generate semantically incorrect SQL code due to their inability to handle complex queries and lack of business context. Dataherald, an open-source NL-to-SQL engine, aims to address these issues by allowing developers to fine-tune LLMs to their schema and monitor performance through a UI. The engine uses two LangChain agents for NL-to-SQL translation: the RAG agent for scenarios with limited golden SQL pairs and the more advanced agent that utilizes a fine-tuned LLM-as-a-tool. Dataherald is currently being used by companies ranging from startups to Fortune 500 corporations to power conversational interfaces and enable self-service data access.