Kogi is an conversational AI learning assistant system that is seamlessly integrated with data science learning environments such as Google Colab and is designed to provide question answering and advice on behalf of teachers and Teaching Assistants. Python is easy to learn with its simple syntax and is widely used in education of data science studies at overseas universities. However, because it is developed by an overseas community, many students stumble because they cannot read error messages unless they are fluent in English. We do not simply consider such students’ difficulties as an English capability, but rather as a linguistic gap between the Japanese and Python. Our proposal is to apply a large-scale pretrained language generation model to realize a learning assistant AI that explains error messages in Japanese and conversely presents code when students state their intentions in Japanese. Ultimately, we would like to provide an environment in which anyone can learn programming and data science in Japanese without any inconvenience, thereby helping to eliminate teacher shortages and regional disparities, and to develop highly skilled digital human resources.
