It’s difficult to generate the code for a whole consumer interface utilizing a Giant Language Mannequin (LLM). Person interfaces are complicated and their implementations typically encompass a number of, inter-related recordsdata that collectively specify the contents of every display, the navigation flows between the screens, and the information mannequin used all through the appliance. It’s difficult to craft a single immediate for an LLM that comprises sufficient element to generate a whole consumer interface, and even then the result’s continuously a single giant and obscure file that comprises the entire generated screens. On this paper, we introduce Athena, a prototype software era setting that demonstrates how using shared intermediate representations, together with an app storyboard, information mannequin, and GUI skeletons, might help a developer work with an LLM in an iterative style to craft a whole consumer interface. These intermediate representations additionally scaffold the LLM’s code era course of, producing organized and structured code in a number of recordsdata whereas limiting errors. We evaluated Athena with a consumer examine that discovered 75% of members most popular our prototype over a typical chatbot-style baseline for prototyping apps.
