Hackathons have gained traction as a compelling educational tool in data science and AI, yet evidence for their long-term effectiveness remains sparse. This paper investigates a structured hackathon curriculum, focusing on a medium-scale program in Thailand that enrolled 175 participants over two months. Seven weeks of Kaggle-style challenges were integrated into an intensive structure, blending online and with onsite sessions. Our results indicate sizeable learning gains: participants’ mean exam scores improved post-program, and hackathon metrics proved to be moderate predictors of formal assessment outcomes. Demographic factors had minimal impact on performance, whereas behaviors such as rapid iteration and collaborative problem-solving stood out. Qualitative data further highlight how hackathons can develop critical soft skills, including leadership, adaptability, and communication. While questions of fairness and inclusivity persist, our findings suggest that well-orchestrated hackathons can enrich AI education by targeting industry-relevant competencies often underassessed in traditional exams.