New Agentic Framework Automates Context-Aware Data Quality Assessment.

Data quality assessment is a critical but challenging task, often limited by its context-dependent nature and reliance on static rules or manual efforts. While large language models (LLMs) offer potential for automation, concerns about reliability and execution safety persist. This paper introduces a unified agentic-retrieval framework designed for autonomous, context-aware data quality assessment. The framework interprets natural-language descriptions of how data will be used, then employs a multi-agent workflow to derive context-specific assessment strategies and generate executable validation logic. To ensure operational reliability, the framework includes a feasibility validation stage. This stage evaluates the realism and executability of the generated assessment specifications, allowing for iterative refinement before execution. Once accepted, the validation logic runs deterministically, guaranteeing reproducible and auditable results. An end-to-end prototype demonstrated that assessment outcomes adapt meaningfully to different intended uses, while the feasibility gate reduced unrealistic rule generation.