Abstract:

The Complete Artifact Workflow (CAW) has recently been proposed as a methodology that enables non-technical stakeholders to independently generate standalone educational web applications using generative AI. Initial studies demonstrated its effectiveness in producing functional user interfaces and static learning modules, such as quiz-based activities and calendar learning tools designed for children with special educational needs.

This paper explores the application of the CAW methodology to a more complex domain involving not only geometry and shape recognition tasks but also freehand shape drawing and automated shape detection. While the generative AI chatbot was able to rapidly produce the graphical interface, static learning tasks, and a functional canvas-based drawing environment, it struggled to generate reliable algorithms for recognizing freehand-drawn shapes. The initial solutions performed poorly, frequently misclassifying the imprecise drawings typical of children and occasionally even misidentifying accurately drawn shapes.

Improving the system therefore required an iterative interaction process in which the operator repeatedly evaluated and criticized the recognition results produced by the AI. Although the AI continuously proposed revised implementations, many intermediate solutions remained inadequate or introduced new errors. Only after a substantial number of conversational iterations did the system eventually produce a functionally robust shape-recognition mechanism.

The final artifact demonstrates that advanced interactive educational tools can be created without the user manually writing a single line of program code. However, the extended troubleshooting process observed in this study highlights an important practical limitation of the CAW methodology. It should also be noted that the requested functionality represents a comparatively complex requirement for a fully standalone offline web application. Implementing robust freehand shape recognition under such constraints involves non-trivial algorithmic challenges. Consequently, the partial difficulties encountered when applying the CAW methodology in this context are not entirely unexpected. At the same time, such advanced interactive functionalities constitute an important category of tailor-made assistive learning tools designed to support the specific developmental needs of children with special educational needs. It remains uncertain whether parents or educators with minimal technical experience would possess the persistence and communication skills required to guide the AI through such a prolonged iterative refinement process in order to achieve satisfactory performance.

Keywords:

Complete Artifact Workflow, Shape Recognition, Geometry Learning, Generative AI, Special Educational Needs, Iterative Prompting.