The teaching experience gathered in the Ambient Intelligence course, that is based on problem-based learning with multisciplinary student groups, has been published in the prestigious IEEE Transactions on Education. The paper focuses on the quantitative and qualitative evaluation of the benefits brought by the teaching methodology, and analyzes three specific research questions: 1. Can the complexity of an AmI system be tackled by third-year undergraduates with limited background knowledge? 2. Can students, design, prototype, and demonstrate a complete project within a single semester, and with their limited experience? 3. Is the collaboration of students from different back-grounds fruitful in AmI systems development? The paper presents the learning goals and teaching strategies, analyzes the learning outcomes from the qualitative and quantitative points of view, and highlights the lessons learned in the process.