In the “DataGenie” project, the team is developing a functional prototype for AI-powered data quality analysis of product master data. A chatbot gathers domain-specific contextual knowledge from data owners and employees in the business departments. This enables large language models (LLMs) to account for company-specific rules, detect errors more accurately and improve positive predictive value.
LLM-based data quality analyses typically rely on general knowledge. Company-specific rules, processes and nuances, however, often reside only in the minds of employees or in documents within isolated data silos. This situation frequently leads to a high number of false positives while actual errors remain undetected. The key challenge is to reliably capture relevant contextual knowledge in the chat as completely as possible. To do this, static baseline questions must be combined with dynamic questions that arise from anomalies in the data.
The project develops a chat agent for context gathering in accordance with standard agent conventions. To this end, the team creates a static questionnaire with follow-up questions as well as a system for dynamic question generation. This system clusters product master data and analyzes it using classic algorithms to identify anomalies, thereby deriving targeted follow-up questions. In addition, the system integrates an LLM-based data quality analysis that detects errors by incorporating the captured contextual knowledge.
By the end of the project, the team will develop a functional prototype that captures contextual knowledge through dialogue, asks static and dynamic questions and uses the answers to analyze product master data. DataGenie is designed to detect more technical errors, reduce false positives and permanently preserve expert knowledge. This solution streamlines the work of data managers when verifying and sustainably improving data quality.
Fraunhofer Institute for Software and Systems Engineering