CLEAN BOT – An Intelligent LLM-Powered Data Quality Agent

Abstract

The increasing volume of structured and unstructured digital data has created significant challenges in maintaining data quality across modern information systems. Conventional data cleaning approaches primarily rely on predefined rules and manual intervention, making them less effective in handling semantic inconsistencies, missing values, duplicate records, and heterogeneous data formats. This project presents Clean Bot, an intelligent data quality agent that utilizes locally deployed Large Language Models (LLMs) through Ollama to automate data cleaning and structuring tasks. The system integrates a Streamlit-based interface with a modular architecture to support data preprocessing, anomaly detection, validation, transformation, and structured output generation. By applying prompt-based reasoning, the proposed approach identifies inconsistencies, removes irrelevant information, standardizes data, validates fields, and converts processed information into structured JSON format. The modular design enables efficient handling of multiple data sources while supporting scalable and maintainable data quality management. Experimental implementation demonstrates that the proposed system provides an effective solution for improving data consistency, accuracy, and organization, making it suitable for intelligent data processing in real-world applications. Keywords— Data Quality Management, Data Cleaning, Large Language Models (LLMs), Ollama, Mistral, LLaMA, Streamlit, Intelligent Data Quality Agent, Prompt Engineering, Anomaly Detection, Data Validation, Data Transformation, Information Extraction, Structured JSON Output, Unstructured Data Processing.