Data analytics developer AtScale claims to have set a new standard in Text-to-SQL accuracy by combining its Semantic Layer and Query Engine with large language models (LLMs) to achieve 92.5% accuracy across all combinations of question and schema complexities.

As enterprises generate and store increasing volumes of data, the demand for quick, accurate data analysis has never been higher, outpacing traditional methods reliant on human analysts. AtScale's integration of Generative AI transforms natural language queries into precise SQL commands, improving efficiency and decision-making speed.

While LLMs excel at generating human-like text, they often struggle with complex database schemas and business logic. AtScale’s Semantic Layer bridges this gap by providing LLMs with comprehensive business-side metadata, eliminating the need to create metrics from scratch or generate complex joins, and significantly enhancing result consistency and accuracy.

“Our integration of AtScale’s Semantic Layer and Query Engine with LLMs marks a significant milestone in NLP and data analytics,” said David Mariani, CTO and Co-Founder of AtScale.

“By feeding the LLM with relevant business context, we can achieve a level of accuracy previously unattainable, making Text-to-SQL solutions trusted in everyday business use.”

In rigorous testing, AtScale’s integrated solution outperformed traditional methods by a wide margin. Across a diverse set of 40 business-related questions, the solution achieved a 92.5% accuracy rate, compared to just 20% for systems without the Semantic Layer. These results underscore the system's capability to handle a wide range of query complexities with superior precision.

Key Benefits of AtScale’s Solution:

1. Enhanced Accuracy: Achieves 92.5% accuracy in translating natural language questions into SQL queries.
2. Simplified Query Generation: Removes the need for LLMs to generate joins or complex business logic, reducing errors and improving efficiency.
3. Business Context Integration: Provides LLMs with essential business metadata, ensuring consistent and accurate results.

The company plans to enhance the integration further by optimizing prompt engineering and expanding training datasets, aiming to tackle even more complex queries with greater precision and efficiency. By doing so, AtScale seeks to empower businesses with increasingly robust and reliable data analysis tools.

You can download the full report here, which includes detailed statistics, charts, and a deeper dive into the analysis.

https://www.atscale.com/