Accura: A Web Application for Detecting StructuralInconsistencies in Migration Datasets Using GPT-4o

Abstract

Data migration in the InsurTech sector requires reliable validation processes to ensure the quality, consistency, and structure of information before loading into enterprise systems. Traditional methods based on manual review, spreadsheets, and SQL queries are often slow, error-prone, and dependent on specialized staff. This study proposes Accura, a GPT-4o-based web application that interprets natural-language validation rules and converts them into executable structures to detect structural inconsistencies in migration datasets. The methodology involved a case study with four representative datasets: policies, issue orders, customers, and invoices, totaling approximately 400,779 records. Results show that Accura achieved 97.14% rule definition accuracy, reduced average validation time from 183.75 to 28 minutes per template, and achieved 100.00% precision, 99.83% recall, and 99.92% F1 score in error detection. Additionally, Accura facilitates the participation of technical and non-technical users while enhancing traceability. In conclusion, Accura improves the efficiency, reliability, and transparency of structural data validation processes in InsurTech migrations.