Bridging Users and Data: Integrating LLMs and CDCS for Trusted, Data-Grounded Answers

Overview

Research datasets can be difficult to access and use, often requiring specialized query skills or familiarity with specific database systems. These challenges erect barriers between the data and the people who need it most—researchers, analysts, and decision-makers who may not have technical database expertise.

NIST researchers are addressing this challenge by building a system that lets users interact with research data in plain English and receive answers grounded in real, trusted sources. The system connects large language models Large Language Models (LLMs) to the Configurable Data Curation System (CDCS), an informatics platform for managing and sharing structured research data developed by NIST.

Approach

The project provides a bridge between LLMs and the CDCS, a curation system that processes raw data into structured data for AI ingestion. Using the Model Context Protocol (MCP), the system enables an LLM to dynamically query and retrieve the most current information directly from CDCS-hosted datasets, ensuring that responses are grounded in live, accurate data sources rather than relying on the model’s general training knowledge.

The CDCS functions as a secure, modular platform that allows organizations to integrate their existing software, automate workflows, and link multiple databases together for unified search. By leveraging LLMs, this project aims to create a natural “conversational interface” where users can query trusted datasets using everyday language and instantly retrieve accurate information.

This delivers two benefits: it gives the LLM a direct, connection to trusted data sources so its answers are grounded in verified information, and it gives users an intuitive way to explore data without needing to learn specialized query languages or database tools.

Impact

Beyond serving as a practical tool for data access, this project provides a platform for studying fundamental questions about how AI systems interact with trusted but heterogeneous data sources. The team is researching new metrics to investigate the degree, nature, and quality of accuracy, groundedness, and realism that intelligent systems can provide when responding to user queries.

A key technique being explored is Retrieval Augmented Generation (RAG), which retrieves relevant information from trusted sources before generating a response. The CDCS platform provides a basis for exploring this and other emerging approaches over time such as agentic retrieval workflows, serving both as a tool for delivering solutions to research users and as a testbed for studying the properties of these techniques themselves.

Related Projects

Configurable Data Curation System (CDCS)