­­Integrating AI learning into institutions: Lessons learned from the Budapest University of Technology and Economics 

The integration of cutting-edge AI education into traditional academic programmes is a challenge many universities face. It’s a fine balancing act between innovation with accreditation, theory with practice, and flexibility with structure. At the Budapest University of Technology and Economics (BME), the Faculty of Electrical Engineering and Informatics took on this challenge headfirst by piloting the ARISA project within its MSc in Computer Engineering programme. What results is a successful, scalable model for embedding industry-relevant AI skills into formal higher education. 

In this interview, Dr. Bálint Gyires-Tóth at BME develops on the university’s journey from planning to implementation and refinement. How did BME align ARISA’s modular learning outcomes with its existing curriculum? What lessons emerged? How can the experience be replicated for institutions ready to integrate AI learning in their courses/training? With over 400 students participating across two pilot rounds, BME’s story offers valuable insights into the scalability and effectiveness of integrating industry-relevant AI competencies into formal academic programmes. 

What was the process of implementing ARISA into MA courses? 

At the Budapest University of Technology and Economics (BME), the integration took place at the Faculty of Electrical Engineering and Informatics, within the MSc in Computer Engineering programme. Because we knew from the outset of ARISA that we would act as a piloting partner, we were able to plan the integration well in advance. As new courses were submitted for formal accreditation, we shaped their curricula in line with the current state of the ARISA work — first the needs analysis, then the AI strategy, and subsequently the curriculum design phase — so that each accreditation cycle reflected the most recent project outputs. 

In other cases, the accreditation timeline could no longer accommodate full structural changes. For those courses, we made adjustments within the limits permitted by university regulations, revising the syllabus of individual subjects so that their content and learning outcomes became aligned with the ARISA objectives. This dual-track approach — formal accreditation where time allowed, and permitted syllabus revisions where it did not — gave us the flexibility to bring ARISA into the programme for the time of the pilots. 

Who was involved in this process? 

The implementation involved a team of ten trainers, working with students enrolled in the MSc programme. We ran two pilot rounds, in autumn 2024 and autumn 2025, with more than 400 registered MSc students participating across the two cohorts. The student body comprised both Hungarian and international participants. Completion rates ranged between 61% and 96% across the different courses, reflecting both the diversity of the subject matter and the heterogeneous study patterns typical of MSc-level engineering students, many of whom combine their studies with professional work. 

Which parts of ARISA were integrated? 

We implemented selected components of the Machine Learning Engineer EQF 7 programme. The modules covered were Deep Learning Basics, Deep Learning Advanced, and Applied AI Models in Practice. The corresponding learning units were Deep Learning and AI Applications. 

The learning outcomes addressed across these courses included Deep Learning (EQF 7), AI Technologies (EQF 7), Machine Learning (EQF 7), ML Ops (EQF 7), Generative AI (EQF 7), AI Awareness (EQF 6) and Soft Skills (EQF 6). Three of the courses were delivered in Hungarian and one in English, accounting for a combined total of 17 ECTS credits. 

What was the feedback received from teachers and students? 

Trainers used coding exercises as a form of work-based learning, although student engagement varied. Although the exercises were designed to be completed during a single class week, many students chose to leave the classroom early and finish the tasks at home. When an assignment appeared too extensive to be completed within a single session, some students were reluctant to begin at all, which later resulted in a noticeable increase in questions during the home-based phase. By contrast, exercises that produced immediate, visible results — for instance visual outputs or performance metrics tracked on user interfaces— recorded significantly higher completion rates. This pattern suggested that students are more motivated by tasks that yield tangible and quickly observable feedback, with the rapid iteration loop reinforcing the underlying learning. 

To strengthen engagement and learning effectiveness, the trainers drew on a broad range of instructional methods. Blended learning approaches were applied in selected lectures to combine different modalities. Industry professionals from well-known companies were invited to deliver guest lectures, giving students direct exposure to real-world applications and expectations. Project work was a major component of every course, with mid-semester and end-of-semester presentations where students demonstrated applied knowledge and received structured feedback. Homework was evaluated through an automated pipeline built on GitHub Classroom, which enabled efficient, transparent and objective grading. In addition, smaller competitive assignments were introduced to drive motivation: in one case, students were asked to write source code that generated an AI image, with the best submission — selected by peer voting — receiving a small prize. These competitions turned routine assignments into interactive and rewarding exercises. 

Student feedback added valuable qualitative insight. Some students felt the courses covered too much material and would have preferred a narrower scope with deeper treatment of fewer topics. A recurring observation was that students preferred completing practical exercises at home rather than during class, suggesting that the structure and pacing of hands-on activities can be further optimised. At the same time, students clearly appreciated the practical exercises and project-based components, which they considered directly relevant to their professional development. 

The pilot also surfaced several areas for improvement. Trainers observed a gradual decline in attendance and active participation as the semester progressed. One contributing factor was that a substantial proportion of MSc students were employed in parallel with their studies, often working between 20 and 40 hours per week, which limited their capacity to attend classes regularly. A further drop in participation was observed around the mid-term and assessment periods of other subjects, when students understandably prioritised mandatory examinations elsewhere. 

Following the first pilot round, structured interviews with the trainers identified a number of forward-looking improvements, captured here in their original wording: 

“The trainers would implement a more robust flipped classroom or blended learning model. This approach would involve providing all necessary learning materials electronically in advance, allowing in-class time to be dedicated to supporting students in achieving key learning outcomes.” 

“A greater emphasis would be placed on highlighting and explaining the available resources to ensure students are aware of and utilise all provided support materials.” 

“The trainers suggested introducing a test-based examination. This would help standardise assessment and streamline the grading process, providing a more consistent and efficient method of evaluating student performance. The trainers also suggested removing parts (e.g. hands-on exercises) that can be solved with LLMs.” 

These recommendations were addressed and implemented in the second pilot round, and the impact was clearly positive. The refined flipped and blended learning design, the clearer assessment structure and the more standardised testing approach together contributed to higher engagement, more transparent evaluation and a tighter alignment between learning outcomes, teaching methods and assessment. Alongside these changes, the second round also introduced targeted incentive mechanisms to lift in-class engagement: active participation during lectures, the completion of Slido-based in-class questionnaires and the points earned through these activities all contributed bonus points to the final examination grade. This adjustment produced visibly higher engagement during contact hours. 

What are some takeaways from this experience? 

The BME pilots indicate that the Machine Learning Engineer EQF 7 ARISA Learning Programme can function as a scalable, practice-oriented AI education model within formal higher education. The dual implementation pathway — accreditation for new  courses and permitted syllabus revisions for existing ones — has proved to be a practical template that other universities operating under similar regulatory constraints can replicate without waiting for a complete accreditation cycle. The pilot evidence also points to concrete opportunities for micro-credentials, given the modular structure of the ARISA learning outcomes and the strong fit with the realities of working MSc students balancing employment and study. We see continuous, evidence-based refinement — informed directly by trainer interviews and student feedback — as a defining feature of the model and one of the reasons it can be transferred to other institutional contexts with confidence. 

The ARISA project’s integration at BME serves as a compelling case study in the successful adoption of innovative educational models within traditional academic structures. By employing a dual-track implementation strategy, comprising of full accreditation for new courses and targeted syllabus revisions for existing ones, BME demonstrated the feasibility of embedding AI education into formal degree programmes. 

For universities considering the integration of AI-focused curricula, BME’s experience highlights the importance of flexibility for working students, and the value of continuous, evidence-based refinement. Similar approaches can thus be adopted, provided institutions remain responsive to feedback and committed to iterative improvement. 

The integration of ARISA courses at BME bridges the gap between academic theory and industry practice while also setting a precedent for how higher education can evolve to meet the demands of a rapidly changing technological landscape.