From Natural Language to Insights: A Scalable Architecture for Query Transformation and Result Summarization

Abstract

With the rise of big data, there is a growing need for systems that allow users to query large datasets using natural language. This paper introduces a scalable, three-stage architecture that translates user queries into structured searches and delivers concise, meaningful results. The pipeline includes: (i) converting natural language input into structured queries via prompt engineering; (ii) executing these queries on OpenSearch over a large news dataset; and (iii) grouping the most relevant results and then summarizing them using transformer-based NLP models. Although built on OpenSearch, the architecture is compatible with other database platforms such as PostgreSQL, MongoDB, Elasticsearch, and Apache Druid. This design improves usability by making information retrieval more natural, accurate, and scalable. All associated code, dataset references, prompts, and demonstration materials are available at: nl2insights.github.io © 2025 Elsevier B.V., All rights reserved.