DATACANVAS-EDU: AN AGENTIC FRAMEWORK FOR INSTRUCTOR-GUIDED SYNTHETIC DATA GENERATION IN BUSINESS ANALYTICS EDUCATION

Business analytics education often struggles with the trade-off between using real-world data—which can be messy, confidential, or difficult to obtain—and public datasets, which are often overused and prone to training data contamination in large language models (LLMs). DataCanvas-EDU addresses this by providing an agentic framework that allows instructors to design custom, synthetic datasets tailored to specific learning objectives.

DataCanvas-EDU operates through a four-phase workflow: Plan, Create, Verify/Test Analysis, and Evaluate. Instructors interact with an AI agent to define the business domain, student level, and specific patterns they want students to discover. The agent then generates the data, creates reference solutions, and prepares rubrics. This process ensures that the data is built specifically to support the intended learning outcomes, such as identifying group differences, trends, or interactions.

By generating data from scratch, DataCanvas-EDU provides students with unique, unfamiliar datasets that prevent them from relying on pre-existing tutorials or memorized analyses. This encourages students to actively collaborate with AI to investigate problems, rather than simply retrieving known answers. The framework is packaged as a reusable AI Agent Skill, making it accessible for integration into existing agent environments and significantly reducing the manual effort required for instructors to prepare and revise high-quality teaching materials.