Scholarlink

<p>Generative AI is increasingly used in financial analytics to interpret large datasets and support time-sensitive decision making. However, access to financial databases still depends heavily on structured query language (SQL), which limits usability for non-technical users. Existing natural-language interfaces often perform well on simple requests but degrade on complex, nested, or domain-specific financial queries, especially under real-time constraints. This paper presents Kestrel AI, a natural-language-driven analytics system that combines advanced NLP with retrieval-augmented generation to produce executable SQL and corresponding visualizations. The system is designed for large-scale financial data, concurrent workloads, and low-latency execution through a modular architecture with GPU-accelerated inference, parallel retrieval, and caching. Experimental evaluation across multiple financial datasets reports an average SQL accuracy of 92%. The system achieves an average model inference latency of approximately 0.3&#xa0;s, while end-to-end query execution time ranges from 1.2 to 4.7&#xa0;s depending on query complexity. User studies with technical and non-technical participants show higher task completion and lower interaction time relative to selected baseline tools. Overall, the results indicate that retrieval-grounded generation combined with performance-aware system design can improve both accessibility and responsiveness in financial analytics workflows. The implementation and evaluation resources are publicly available at <a href="https://github.com/veerababulara/Natural-Language-Driven-Data-Visualization-Using-Kestrel-AI.git">https://github.com/veerababulara/Natural-Language-Driven-Data-Visualization-Using-Kestrel-AI.git</a> and archived with DOI <a href="https://doi.org/10.5281/zenodo.19642245">https://doi.org/10.5281/zenodo.19642245</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Natural language-driven data visualization using Kestrel AI: a novel algorithm for intelligent analytics

  • 期刊论文

  • 发表日期：

    2026年5月8日

查看全文
  • Veerababu Reddy,
  • N. Veeranjaneyulu

摘要

Generative AI is increasingly used in financial analytics to interpret large datasets and support time-sensitive decision making. However, access to financial databases still depends heavily on structured query language (SQL), which limits usability for non-technical users. Existing natural-language interfaces often perform well on simple requests but degrade on complex, nested, or domain-specific financial queries, especially under real-time constraints. This paper presents Kestrel AI, a natural-language-driven analytics system that combines advanced NLP with retrieval-augmented generation to produce executable SQL and corresponding visualizations. The system is designed for large-scale financial data, concurrent workloads, and low-latency execution through a modular architecture with GPU-accelerated inference, parallel retrieval, and caching. Experimental evaluation across multiple financial datasets reports an average SQL accuracy of 92%. The system achieves an average model inference latency of approximately 0.3 s, while end-to-end query execution time ranges from 1.2 to 4.7 s depending on query complexity. User studies with technical and non-technical participants show higher task completion and lower interaction time relative to selected baseline tools. Overall, the results indicate that retrieval-grounded generation combined with performance-aware system design can improve both accessibility and responsiveness in financial analytics workflows. The implementation and evaluation resources are publicly available at https://github.com/veerababulara/Natural-Language-Driven-Data-Visualization-Using-Kestrel-AI.git and archived with DOI https://doi.org/10.5281/zenodo.19642245.