DataForge: An AI-Driven Data Warehouse Schema Generator

Abstract

Organizations today rely on robust data warehouses to integrate and analyze massive volumes of data for decision-making. However, designing an optimized warehouse schema (fact and dimension tables, keys, constraints, naming conventions) is a labor-intensive, error-prone process that can take weeks of manual effort by experts. This paper presents DataForge, an AI-driven framework that automates and accelerates data warehouse schema creation. DataForge parses input SQL schema definitions, infers an appropriate star schema, and enhances it using AI. It combines regex-based and grammar-based SQL parsing to reliably extract tables, columns, and relationships; keyword-driven domain detection and NLP techniques to infer business context; semantic validation to enforce logical consistency and naming standards; and heuristic rules to classify tables into fact or dimension roles. An interactive web-based interface allows users to visualize and refine the generated schema with real-time suggestions. In benchmark evaluations on retail, healthcare, and financial datasets, DataForge reduced schema design time by over 80 % and achieved an expert-validated schema quality score above 90 %. The results indicate that DataForge can dramatically streamline the schema design phase, paving the way for faster, more consistent data engineering workflows.