AI-Powered Adaptive Study for Self-Regulated Learning: Preliminary Evidence from Ulern in LatinAmerica

Authors: Josefa Karmy, Tomás Berríos, and Fernanda Benavides.

Background: Students face a challenging ecosystem marked by structural and pedagogical barriers that impact their well-being and learning. Evidence shows that major academic problems stem not from intellectual capacity but from ineffective methodologies and other interrelated factors, especially affecting students from historically excluded groups (Putri, 2024). In this context, AI-based solutions are emerging to promote active, adaptive, and scalable learning. For example, Henkel et al. (2024) demonstrated that a low-cost, scalable AI tutor can significantly enhance learning in low and middle income countries. Building on this, the aim of this study is to evaluate the effectiveness of Ulern’s study session methodology on students' academic performance and self-regulated learning.

Intervention: Ulern is an AI-powered learning app developed in Chile to address the challenges of independent study. Its technology generates structured, evidence-based study sessions adapted to the students needs and context. The student uploads any study material (such as texts, notes, photos, videos, etc.) and the app identifies the learning objectives contained in these materials. Then, the intervention is structured into a progressive study session that includes three stages: diagnosis, to assess initial knowledge; guided practices, with explanations and questions that atomize content focusing on each learning objective; and independent practice, to evaluate final knowledge. Ulern seeks to strengthen self-efficacy, foster intrinsic motivation, and promote self-regulated learning, thereby contributing to improved academic experience and results.

Methodology: A mixed-methods approach was employed, combining pre-post quantitative evaluation with qualitative analysis of user perceptions. Quantitative measures were collected in the diagnostic practice at the beginning of the study session and in the independent practices, at the end of it. The primary outcome of performance was the score obtained for each learning objective in both practices. Analysis considered all completed study sessions conducted by any user in their first month using the app (N = 3,254). Paired t-tests were applied to evaluate progress per learning objective, and effect size was estimated using Cohen's d coefficient (Cohen, 1988). The qualitative analysis was conducted through semi-structured interviews and focus groups with voluntary users (n = 15). Inductive content analysis was performed, which allowed for the collection of qualitative evidence on the perceived effectiveness of Ulern's methodology and its influence on study routines and academic performance.

Results: The sample consists of a total of N = 3,254 students who completed a study session between April 6th and July 18th of 2025. Preliminary quantitative analysis shows statistically significant gains in performance by the end of the study session. On average, students increase their content mastery by 27% after their first study session with Ulern. Paired t-tests show a statistically significant change in students outcome (t = 56.10, p < 0.001). The estimated effect size (Cohen d) was 0.42, which is considered a medium to large effect in the context of educational interventions. In the qualitative analysis, central categories related to perceived effectiveness, motivation, self-regulation, and usability were identified. Ulern is highlighted as an active learning solution that structures study, adapts to the student's level and content, and adequately prepares for evaluations. The intuitive interface, engaging resources, immediate feedback, and the progressive structure of the study sessions are all highly valued.

Conclusion: Preliminary evidence suggests that Ulern’s AI-powered, three-stage study model effectively enhances self-regulated learning and performance in a Latin American context. Further controlled trials and longitudinal research are warranted to confirm sustained impacts and to refine the adaptive algorithms for diverse learner profiles.

Keywords: adaptive learning, self-regulated learning, AI-based interventions, mixed-methods, educational innovation, LMIC

References

Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences (2.a ed.). Routledge. https://doi.org/10.4324/9780203771587

Henkel, O., Horne-Robinson, H., Kozhakhmetova, N., & Lee, A. (2024). Effective and Scalable Math Support: Experimental Evidence on the Impact of an AI-Math Tutor in Ghana. En A. M. Olney, I.-A. Chounta, Z. Liu, O. C. Santos, & I. I. Bittencourt (Eds.), Artificial Intelligence in Education. Posters and Late Breaking Results, Workshops and Tutorials, Industry and Innovation Tracks, Practitioners, Doctoral Consortium and Blue Sky (Vol. 2150, pp. 373-381). Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-64315-6_34

Putri, G. A. (2024). Social Support and Educational Resilience: A Systematic Review of Students Facing Academic Challenges. Vifada Journal of Education, 2(2), 24-44. https://doi.org/10.70184/hnxrcx44