Abstract : This paper presents a platform that allows students to generate mobile applications for smartphones and tablets using natural language descriptions. We explore three methods of modifying the model’s input (prompts) to optimize the generated apps. We evaluate the model’s performance using the BLEU score and found that appropriate example pair selection and variation of the number of example pairs improve the quality of the generated apps. Finally, we discuss the implications for computer science education in light of generative models for code.
Download paper here