The lack of prior knowledge in the initial years of professional careers worsened after the COVID-19 pandemic restrictions. Classical theories of learning emphasize the cruciality of previous knowledge to facilitate the understanding of new concepts. Wood et al. (1976) stated that students require structured support to build on their existing knowledge; however, when this prior knowledge is insufficient, the scaffolding process intensifies and becomes more complex.

Learning is most effective when instruction is calibrated just beyond students’ current level, provided there is sufficient prior knowledge on which to build (Vygotsky, 1978). In the absence of such prior knowledge, remedial instruction is required to enable learning. Sweller (1988) indicated that the lack of prior knowledge increases the cognitive load, which hinders the learning process. This idea aligns with Vygotsky's theory regarding the need to apply leveling strategies to support students.

The meta-analysis carried out by Hattie (2008) compiled more than 800 studies on the factors that influence student learning, highlighting that prior knowledge is fundamental because it provides the base on which new learning is built. Students with relevant background knowledge tend to have a deeper and more effective understanding of new concepts. In fact, Hattie (2017) found that prior ability has an effect size of 0.94 on academic performance, which is considered highly significant.

Research shows that engineering students with deficiencies in algebra or calculus are at higher risk of dropout (Klingbeil et al., 2007; McKenna et al., 2000). Drawing on cognitive load theory, Sweller et al. (1998) argue that insufficient prior knowledge can overwhelm students when they encounter complex topics (e.g., design or programming). Therefore, at the university level, bridging strategies—such as remedial courses, reinforcement programs, tutoring, and introductory modules—together with pedagogical approaches that scaffold learning, are needed to ensure that students master the necessary prerequisites before progressing in their programs (Felder & Brent, 2005; Seymour & Hewitt, 1997).

Some universities have adopted multiple strategies to support the academic leveling of new students. For example, California State University has a peer-mentoring STEM program in which students in advanced semesters mentor first-time students, providing them with information about careers and sharing their personal experiences. Having a peer who offers guidance and support during the initial semesters has proven to be an effective strategy to improve retention rates and facilitate a faster, more positive integration into university life (Taha et al., 2015; Rockinson-Szapkiw et al., 2021).

Other universities, such as Massachusetts Institute of Technology (MIT) with its MIT Introduction to Technology, Engineering, and Science (MITES) program or the National Autonomous University of Mexico (UNAM) and its program inspiring women to enter STEM, among many others, have implemented specific programs to strengthen STEM careers. These programs not only offer academic advice to help students level up but also provide emotional support to those who begin their university education, to strengthen their knowledge, and prepare them to face the challenges of these disciplines.

Adding to the structural factors that explain the lack of prior knowledge in particular areas, the COVID-19 pandemic significantly exacerbated this problem (Filatova et al., 2023; López del Puerto et al., 2021). In response, several institutions developed remote programs during and after the pandemic, with a special focus on supporting academic leveling, particularly in STEM courses (Lasater et al., 2021; Schaal et al., 2021).

However, some of the traditional strategies used for leveling, such as preparatory courses before the formal start of the program's subjects, need to be rethought. In the study carried out by González-Beltrán et al. (2018), data mining techniques were applied to analyze whether these leveling courses significantly affected the performance or approval of students in Calculus subjects. However, the results showed no evidence to support such a correlation.

Although leveling courses before university admission have been one of the most common strategies implemented by universities, they are not always the most effective. One of the primary reasons is that these courses tend to address all the contents of the previous educational level in a general way, without considering that students do not all have the same knowledge gaps. This can demotivate students and lead to their negative perception of the usefulness of the content. However, thanks to technological advances, Adaptive Learning (AL) emerges as a promising alternative to achieve academic leveling as it allows the learning process to be personalized, focusing only on the topics that each student needs to reinforce (du Plooy et al., 2024; Taylor et al., 2021).

AL has been implemented for several years, classified under the umbrella of Personalized Learning, according to Jing et al. (2023). The AL approach aims to personalize learning and materialize the principle of student-centered education. Similarly, Hernandez et al. (2022) define Personalized Learning as an educational approach that adjusts the learning experience of each student according to their individual needs, strengths, abilities, and interests, being flexible regarding what, when, how, and where to learn throughout their curriculum and various formative experiences. In this way, multiple strategies and technologies provide different degrees of autonomy and options, allowing students to play an active, leading role in their learning within these environments.

Digital technologies enable multiple approaches to personalize learning. The broader and more thoughtful their use, the more precise and responsive personalization can become. Adaptive Learning is an educational strategy that utilizes data-driven technology to tailor learning itineraries for personalized learning, providing effective and efficient content pre-designed by the teacher based on the student's performance level, profile, and learning needs. This allows teachers to identify gaps in content comprehension to improve actions and adjust their educational practice, optimizing student performance (Rincon-Flores et al., 2024). Thus, the teacher’s role shifts from content provider to facilitator of learning. However, to achieve these efficiencies, teachers must use technological systems appropriately and efficiently (Cavanagh et al., 2020).

Based on a computer science and information technology survey in higher education, Simon and Zeng (2024) agreed that most university leaders and faculty display a positive attitude towards Adaptive Learning (AL) and consider it a strategy with high potential to improve student success. In addition, Rincon-Flores et al. (2023) demonstrated that technology-assisted AL effectively optimizes learning. Similarly, Sarıyalçınkaya et al. (2021) argue that AL allows the teaching trajectory of each student to be defined according to their learning style, based on the evaluation of their previous knowledge.

In the field of technological systems, one must understand the relevance of data and its processing. Long and Siemens (2011) highlighted that the strong drive to increase efficiency and reform higher education has made it natural to orient learning towards new digital formats. This has led to an explosion of compiled educational data, including activity flows, student interactions with peers and mentors, administrative records, and other valuable sources of information. The possibility of collecting this massive volume of data, analyzing it in real time, and applying machine learning techniques to identify patterns is primarily due to the support of artificial intelligence (Miao et al., 2021). Today, these elements are integrated into adaptive learning platforms that incorporate artificial intelligence (AI) as the core of their operations.

Fernández-Morante et al. (2022) reported that learning analytics has enormous potential to optimize educational processes; however, for this potential to produce meaningful results, the focus must shift from technology to learning-centered instructional design. Similarly, Kara and Sevim (2013) added that adaptive platforms use algorithms, assessments, feedback, adjustments, instructor intervention, and different means to offer new learning materials to students who have reached the expected level or provide reinforcement activities to those who have not yet achieved it.

One of the main advantages of adaptive platforms lies in their ability to generate learning analytics and monitor student progress at both the individual and group levels. These tools collect real-time data on performance, study habits, and areas of opportunity, enabling more timely and effective pedagogical decision-making, both during and after implementation. The strategic use of adaptive platforms with AI integration not only optimizes teaching intervention but also strengthens personalization and promotes equity in STEM courses, characterized by high academic heterogeneity (Koester et al., 2016; Li et al., 2023). These technologies facilitate differentiated learning paths adjusted to the specific needs of each student, thus increasing their chances of academic success.

In this sense, Moskal et al. (2017) highlighted that adaptive platforms employ a data-based and, in some cases, non-linear instructional approach, which promotes academic leveling and continuous content adaptation. This perspective sees adaptive platforms as a high-value pedagogical resource for addressing the challenges of teaching in complex and diverse educational environments.

Regarding instructional design, other methodologies, such as Self-Regulated Learning and Microlearning, can complement AL (Rincon-Flores et al., 2024). The former requires the student to use self-regulation strategies, respond to feedback, and maintain interdependent motivations (Zimmerman, 1990), while the latter offers brief, focused content, facilitating the assimilation of key concepts (Allela et al., 2020). An effective ALS integrates structured content design, relevant teaching strategies, a coherent implementation model, and technologies that enhance personalization (Rincon-Flores et al., 2024; Simon & Zeng, 2024; Shi & Liu, 2025).

A successful example of an ALS application in STEM courses has been at the University of Central Florida (UCF), which redesigned its physics courses by incorporating appropriate technology to serve 4,500 students, achieving improvements in student success, increased learning outcomes, and greater engagement with content (Dubey et al., 2023). Similarly, India’s state of Haryana, aware of AL’s positive impact, became the first to implement it on a large scale, selecting an educational technology partner to provide relevant software and content for 500,000 tablets distributed to public school students across the country (Press Trust of India, 2022).

On the other hand, the results of AL are not always successful or conclusive. An example is the study by Wang et al. (2023), where, although students in the experimental group (adaptive learning) improved significantly between pre-test and post-test, their results were not statistically significant from the control group (teacher-led learning). Similar findings were reported by Eau et al. (2022), who found benefits only in some students, particularly those with high performance. Similarly, Olmos-López et al. (2023) evaluated an approach in which teachers personalized instruction using a predictive model of student performance; however, the predictions were primarily useful for estimating group performance. Thus, instruction was not personalized at the individual level but was adapted to group needs.

The ALS presented in this study integrates adaptive leveling modules within first-year courses in STEM programs, addressing basic knowledge that students should have mastered by the time they graduate from high school. Each module begins with a mandatory diagnostic exam that defines a personalized learning path, allowing only the necessary content to be taken. This occurs in the first few weeks but remains available for consultation throughout the course. Participation is not optional and represents 5% of the final grade, encouraging its use and ensuring that all students reach the required level.

The ALS presents four key components to ensure its pedagogical and technological effectiveness: content design, the didactic model, the role of the teacher, and the technological platform.

Content design is based on micro-learning, breaking content down into small lessons with specific goals. Each lesson includes contextualization, recognition material, examples, exercises, and a final quiz. The knowledge-check resources include one base and two optional reinforcements. The contents are presented in various formats, including videos, infographics, readings, and animations. The quiz has items of varying difficulty depending on the student's level.

Based on principles of self-regulated learning and self-study, the didactic model allows students to control their learning and adapt it to their needs. The pedagogical structure is designed for autonomous learning, managing time, pace, and needs. The student begins with an initial diagnostic examination. Based on the results, each reflects and follows a personalized learning path that indicates lessons to be reinforced. Each lesson includes context, videos, explanations, examples, exercises, and a quick quiz. A final diagnostic exam is presented at the end of all lessons, and the grade is recorded.

Although the self-study model promotes student autonomy, the teacher is key to the student’s success. The teacher motivates the class, provides guidance regarding the adaptive module, and ensures that diagnostic tests are performed in class. The teacher reviews AI-generated outcome analytics on the technology platform to identify and track specific needs, thus acting as a strategic companion in the leveling process.

To implement the ALS, it was necessary to select a platform that would allow teachers to create content, integrate various materials, and use AI for analysis, monitoring, and generation of personalized paths. RealizeIT was selected after the assessment of different options for its integration with the LMS, variety of content formats, detailed analytics, user-friendly interface, and its Adaptive Intelligence Engine. This engine uses mathematical and statistical models and algorithms to record and analyse the progress, strengths, weaknesses, and preferences of each student (Howlin & Lynch, 2014). It detects learning patterns and offers personalized and adaptable paths based on each student's interaction with the content, as well as recommendations for resources and practices to facilitate the mastery of concepts.

To evaluate the impact of the Adaptive Learning Educational Strategy (ALS), implemented through the technological platform, on the leveling of students' prior knowledge in the Mathematical Thinking (MT), Mathematical Reasoning (MR), Mathematical Modeling (MM), and Computational Thinking (CT) training units, and to analyze the learning experiences of teachers and students.

The non-probabilistic sampling process resulted in a sample comprising 1,309 students from the Schools of Engineering and Sciences, School of Business, and School of Social Sciences and Government. Samples per course were matched; for this, the participants of the experimental and control groups with equivalent characteristics were selected using SPSS software (version 29). The variables considered for the pairing were: final high school grade, gender, type of school of origin (public or private), and grade obtained in the pre-test. As a result, the matched sample comprised 734 students: 120 from the Mathematical Thinking course, 50 from Mathematical Reasoning, 410 from Mathematical Modeling, and 154 from Computational Thinking. Figure 1 shows some sociodemographic characteristics of the total sample.

Figure 1 Figure 1 Sociodemographic statistics concerning gender and scholarship status Figure 1 Sociodemographic statistics concerning gender and scholarship status

In addition, 88% of students came from private schools and 12% from public schools. Regarding the teaching staff, 40 teachers participated in both treatments; however, only 11 responded to the final survey. Their average age was 40, and the average teaching experience was 10 years.

A pre- and post-test of previous knowledge was applied in each course. These instruments were created by the teams of teachers in charge of developing the contents of the leveling modules. Subsequently, a pilot study was conducted to verify that item difficulty and discrimination indices were appropriate and that the scales showed adequate internal consistency (Cronbach’s alpha), see Table 1.

Table 1 Table 1 Psychometric results of the instruments

Course	Number of students	Difficulty Index	Discrimination Index	Cronbach’s alpha
Computational Thinking	28	0.71	0.42	0.8
Mathematical Reasoning	36	0.61	0.48	0.86
Mathematical Thinking	41	0.71	0.29	0.78
Mathematical Modeling	34	0.57	0.33	0.74

Each instrument was piloted with twice as many items as planned. Following the analysis, the 16 items with the best psychometric results were selected, resulting in each instrument consisting of 16 items.

At the end of each course, a survey was applied to students and teachers in the control and experimental groups. The objective of this questionnaire was to collect information on the learning experience in relation to the ALS for leveling. The dimensions of this instrument were: Learning Path, Learning Process, Teaching Process, Content, Engagement, Usability, User Experience, and Satisfaction. The instrument utilized a continuous-interval Likert scale with a range of 0 to 100 points to capture nuances in the intensity of each response.

ANOVA tests were performed to analyze the results of the pre- and post-test, after verifying the assumptions of normality and homogeneity of variances. Likewise, a Difference-in-Differences (DiD) analysis was applied to estimate the real effect of the treatments. As for the survey, since it used a continuous Likert-type scale, chi-square tests identified differences between the teachers in the experimental groups. Also, the use of data provided by the adaptive platform was analyzed. Finally, the open questions in the questionnaires were analyzed through thematic categorization to complement and enrich the quantitative results.

The study was carried out in accordance with the ethical principles set out in the Declaration of Helsinki. All students, professors, and academic authorities were informed about the objectives and procedures of the study. Participants were explicitly notified that all information collected would be held to strict confidentiality standards, ensuring absolute anonymity. Moreover, all participants gave their informed consent before starting their participation.