Machine learning (ML) has significantly enhanced applications across various domains, such as computer science, art, medicine, and energy management. This fact has attracted end-users with little to no machine learning expertise to learn how to create ML applications to solve their domain-specific problems. Although ML techniques have become prevalent, it is still challenging and high-barrier for non-expert ML users to learn and build ML models with satisfactory performance. Unlike existing interactive tools that support ML education mainly for visual recognition tasks, we create a prototype tool where ML learners can interactively build an autonomous game agent. The tool starts with the educational purpose, takes the form of a learn-by-doing tutorial, and is designed while considering the experimental and iterative nature of ML. Since the tool design is still in the early stage, we discussed the potential future work, like a hinting feature that helps the user analyze and fix model issues and the relationship with programming-by-example.