Author: Livija Smrkolj, 4th year student at the Diocesan Classical Gymnasium (SLOVENIA)

Contact: livija.smrkolj@stanislav.si

Summary

The integration of artificial intelligence (AI) into education is redefining traditional learning methods, particularly in the areas of personalised learning, AI tutoring and digital education. One innovative AI-driven platform that I know and have tested is Astra AI (released in 2023), an application developed by Slovenian innovator Andrej P. Škraba that supports students in the subjects of math, chemistry, English, German and physics. With over 70,000 users, Astra AI offers an adaptive learning experience and provides instant step-by-step solutions to problems submitted by students via text or image input.

Multilingual Astra AI makes the program accessible to a wider audience, allowing students from different language backgrounds to benefit from its features. The study begins with an overview of Astra AI’s core features, including AI-generated study notes that provide comprehensive summaries of any topic of interest. In addition, the Matura AI feature embedded in Astra AI enables structured preparation for the national Matura exam in Slovenia by guiding students through the curriculum of the above-mentioned subjects. It assists primary school students (pupils) with the national knowledge tests and also provides advanced IB math preparation with expert guidance, over 2,000 previous exam questions and detailed solutions of previous final exams.

One focus of this article is the comparative analysis between AI-driven tutoring and traditional teaching from my personal perspective. The results show that AI tutoring outperforms traditional tutoring in terms of efficiency, accessibility and personalised feedback. Unlike traditional tutoring, it provides instant explanations, customises explanations to the student's learning preferences and is significantly more cost-effective than human tutors. Despite these benefits, I am also exploring its limitations, such as occasional inaccuracies in problem interpretation, misinterpretation of poor quality images uploaded to the app, and the use of alternative solution methods with which students are unfamiliar. These issues are mitigated when students clarify their questions, indicate preferred solution techniques or refine the quality of input.

Empirical data proves the success of Astra AI: 99% of students improved their understanding and academic performance after using the platform, according to the app developer. This study not only assesses the strengths and weaknesses of AI-driven learning, but also discusses its implications for the future of education. The findings contribute to the discussion on the role of AI in personalised education, its potential to complement traditional teaching and its transformative impact on student engagement and independent learning.

In addition to a limited analysis of AI in education, this paper also presents my perspective as a a high school senior by detailing the effectiveness of Astra AI. The discussion includes its real-life applications in exam preparation, concept mastery and independent problem solving, with emphasis on how it supports students in their final year of secondary school.

Astra AI represents an advance in digital education by promoting problem solving, critical thinking and self-directed learning. As AI continues to reshape education, this paper highlights the need for further research on the long-term impact of AI on learning outcomes, its potential integration into traditional curricula and strategies to overcome its current limitations.

Keywords: Artificial intelligence, personalised learning, mathematics education, digital learning tools, AI tutoring, student engagement, AI-generated learning notes, AI-assisted exam preparation

*******************************************************************************************************************

Introduction

The integration of AI into the educational context is no longer a futuristic concept, but an urgent pedagogical reality. Students, teachers and institutions around the world are increasingly turning to digital solutions to improve learning, close knowledge gaps and personalise content delivery. Astra AI, a noteworthy learning platform developed in Slovenia and tailored to subjects such as math (primarily), physics, chemistry, English and German, offers a unique and scalable approach to AI-powered learning.

In this article I would like to offer a critical evaluation of Astra AI, both from a theoretical point of view and from my personal experience as a secondary school student preparing for the national Matura exam in Slovenia. The paper also reflects on broader educational trends related to AI tutoring and highlights practical, ethical and pedagogical considerations for integrating AI into learning environments

The Context of AI-Supported Learning in Math

Math, often considered one of the more abstract and challenging school subjects, has long benefited from structured explanations and repetition. In recent years, AI tools have emerged that offer both. While global platforms such as ChatGPT incorporate AI to varying degrees, region-specific solutions such as Astra AI provide targeted, localised support based on national curricula. In Slovenia, where the Matura exam plays a central role in university admission, tools like Astra AI and eMatura are beginning to change the way students prepare for exams and master complex material.

The Astra AI Platform: Features and Functionalities

Astra AI was launched in 2023 by Slovenian educator and innovator Andrej P. Škraba. Its design is based on the concept of adaptive learning, with an AI backend that analyses user input (text or image-based) and generates customised, step-by-step solutions to mathematical problems. The multilingual user interface and embedded Matura AI module are particularly useful for Slovenian students preparing for national exams. The platform also includes structured video explanations, learning plans and more than 2,000 pre-set exam questions. Features such as AI-generated notes, personalised quizzes and progress tracking provide an integrated experience that is closely aligned with cognitive learning theory and constructivist pedagogy.

Astra UI is designed to bring math closer to every student and drastically reduce the cost of teaching. Instead of expensive one-on-one tutoring, the web and mobile app offers interactive AI learning that focuses on understanding the material, not just finding the right answers. The app offers help in math (for which it was developed), English, physics, chemistry and German as well as learning support for the primary school exams “NPZ” in Slovenia and the International Baccalaureate IB. It also contains video explanations of all the math material (which the developer Andrej P. Škraba also offers for free on YouTube) and a guided plan on how to work through all the math material up to the Abitur, with all the explanations and exercises. The app helped me the most in math, and partly also in chemistry.

Astra AI, a Slovenian paid online application for learning math, is already used by more than 70,000 students across Slovenia, according to the developer, and offers Slovenian math teachers free access to an AI tutor (Siol.net 2024); (AIDEA 2024).

My Personal Learning Experience

The subject of math at school was a bit of a struggle for me at first - I often got stuck on more complex equations and abstract concepts. I had the feeling that "math really wasn't my area". Despite my best efforts in lessons, the material in the last year of grammar school sometimes remained unclear and the whole of the material unconnected. I started to lose confidence and patience. I decided to integrate digital tools such as special learning tools, a learning management system-LMS ("eMatura", "Astra AI" - partly already AI-driven), and the AI app ChatGPT into my daily self-learning. I have found that modern learning tools can offer me additional explanations of the material, tailored to my understanding, and interactive exercises that go beyond the time constraints of a normal class. Learning in this way has opened up a new and unique way of explaining things, and I can now practice math at my own pace, anytime, anywhere. The LMS tool in question, enhanced by the AI, Astra AI, has impressed me, but it is not infallible, and the user has to judge critically the results presented.

I started using Astra AI regularly in autumn 2024, in my 4th year of high school, and it has gradually changed the way I learn math. I wanted to try it out because of the Baccalaureate. Its 2500 free video explanations on YouTube were the trigger for me to try the paid-for GPT-4 Turbo AI-powered app, as I was hoping for a "cheaper version of one-to-one help". It is also optionally designed in the "Socrates" mode, which is interpreted as the search for truth through dialogue and thought, while one matures and mobilises one's resources and reflects on the problems that cause them, which works very much like the help of a "human" coach. The aim of the method is not to give an answer to the questions of others, but to enable that person to deepen his or her reflection and to be able to find a solution, even if invalid, on his or her own. Using it in the "Socratic" mode requires more independence and depth, and I was not impressed. I mostly used the "Solver" option, which quickly answers the question together with an explanation. The "Alpha" mode is no longer available, but it was a bit more of a wacky way of delivering the material and closer to teenagers, using teenage colloquial words in the explanations. As my classmates told me, he made several mistakes and was inaccurate. This is probably why the subject app no longer allows this way of learning.

Astra's AI-driven digital tool is an amazing learning tool that I couldn't have imagined a few years ago. I'm looking forward to it because most of the math teachers at my school and lecturers I know haven't even tried this tool and aren't particularly excited about it either, perhaps because of the potential for misuse and because they don't see any added value in AI/digital tutoring. The Astra AI app is also not yet used by many Slovenian teachers, which is why developers are making it available to teachers free of charge as part of a special campaign in spring 2025 (Svet24 2024).

I tested the usefulness of the tool myself and independently. I am an Honours student and did not have any AI-based learning tools in my first three years at secondary school (as they were not available) and I did not use any other specialised LMS (eMatura). So far, I have not involved any math teachers in my learning. In my fourth year, in the school year 2024/2025, from September 2024 to April 2025, I spent a total of less than €90 on the use of the AI tool, the AI online learning application Astra, which is relatively little compared to the total hours I would have paid for a "human" teacher (one hour of the latter costs about €20).

How I used Astro AI and what I learnt from it

I used the app at least twice a week from September 2024 to April 2025, i.e. an average of 4 hours a week. In times when I was pressed for time, when I had many exams coming up, when I only had the late evening hours available during the day and when there was no one around to understand and master the material, this learning AI proved to be an excellent additional learning tool to supplement the limited pedagogical process of teaching and consolidating the material in the classroom according to the official curriculum. As I was often absent from class during the year and missed math lessons, the Astra AI provided me with excellent support in initially understanding the material and preparing notes on the material and has so far proved its worth in this regard. Most of the time I have asked the app for additional tasks on a particular topic, which I have been able to solve independently or with the help and explanation of the AI. It worked quite well here.

When I couldn't solve a math problem, I took a photo of it with my smartphone and asked the app to explain it. I wrote in the instructions that it would describe all the steps. Up to this point, I always understood the explanations. In my tests with the new tools, the Astra AI app makes fewer errors than the more widely used ChatGPT AI tool and generally uses the correct equations in the calculation. So far in my use, Astra AI has only made errors in interest calculations and integrals. The latter are often calculated incorrectly by the mathematical LMS based on UI tool mentioned above and with different formulas than we are used to from school; here Astra AI falls short of expectations. This is probably due to the incomplete design of the system database in the area of interest calculations. When using such an AI learning application, the user must take special care to critically examine the results provided.

According to research by AI developer Astra, initial feedback from students on their personal learning experiences shows better understanding and consolidation of the material, particularly in math where practise and step-by-step problem solving are crucial (Siol.net 2024). This new concept of AI-mediated learning, I believe, can directly help solve a long-standing problem of traditional teaching, where limited time and diverse backgrounds, as well as oral communication in the classroom, make an individualised approach impossible - unprocessed material accumulates and students quickly lose confidence as the material piles up.

The key for the app in my case was to ask a question that the AI understood - for example, I found that I got a better explanation and a more complete answer from the AI when I asked, "Can you explain the steps to solve this equation again, like you would to a ninth grader?" rather than just "I don't understand." Well-designed AI questions lead to very useful, customised answers, while questions that are too general and not detailed enough provide incomplete information and less informative answers. It is also necessary to delve deeper into the AI instructions. For this, there are already special 'prompt generators' or AI e-instructions that help us to write AI instructions in such a way that we really get what we want as an answer (Feedough 2024). I also recognise that repeated testing of the same questions in an AI tool will show differences and the reliability and relevance of the results must be critically assessed. This happened to me with interest calculations when I had to explain three times exactly what I wanted the answer to be before it solved the problem correctly.

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

Figure 1, 2, 3: Screenshot of an example of solving a photographed task in Astra AI, in Slovenian language only, February 2025

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

Figure 4, 5, 6: Screenshot of an example of the interpretation of the substance integrals in Astra AI, in Slovenian language only, March 2025

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

‍

Figure 7, 8: Showing incorrect answers given by the Astra A app when the app does not understand the instructions given for a math problem, in Slovenian language only, April 2025

I believe that even with the help of the Astra AI learning app and the occasional use of the free version with a registered ChatGPT account, I learnt the material faster and consolidated its quality. Despite the more challenging material in math in the final year of high school, my performance remained at a high level, grade 5. I have excelled in math in all four years of high school, so it is harder to prove the value of the app, but I can certainly confirm that with the help of the e-tools I reduce my explanation and learning time, but the time for consolidation remains the same, because consolidation is the most important factor contributing to my good performance.

Pedagogical Implications and Comparison with Traditional Tutoring

One of the biggest advantages of AI in education is probably the ability to personalise, differentiate and individualise learning for students. While traditional ex cathedra teaching methods make it difficult to fully customise the pace and explanations to each individual, AI can do this dynamically. AI tutors (e.g. the chatbot ChatGPT or the specialised AI Astra) analyse the learner's questions and answers on the fly and adapt to the learner's level of knowledge. If the learner asks the questions correctly, they can receive the explanation like a tutor, in a form that they can understand.

Another advantage of an AI learning app is the personalisation of the tool, which is reflected in the ability for the AI to provide me with a personalised learning plan or set of exercises according to the learner's needs if required - this would be difficult for a teacher in a classroom with more than 30 students

The individualisation of learning through the Learning AI app, an added benefit, means that I can also learn and practise independently and at my own pace without being guided by a teacher or instructor. My personal learning process is individualised.

Differentiation is already guaranteed in secondary schools, because after primary school we are allocated to secondary schools according to our interests and performance. An AI learning app could help here with even more profiled differentiation by tailoring the scope of the material and the way explanations are given even more closely to the individual school programme.

The biggest advantage of the AI tutors for me is the 24/7 availability, because I usually study in the evenings and when I am looking for explanations, the tutors are not available. This availability is further enhanced by the always immediate feedback. While a live tutor is not available outside of class (and even private tutors have limited hours), the AI is always available 24/7. This convenience and accessibility are, also reflected in all the articles already mentioned as expert contributions (Kovač Polajnko, Jereb and Cugelj 2024); AI-assisted learning takes place anytime, anywhere and offers extreme flexibility to students. In addition, both the AI learning application Astra and the classic LMS (e.g. eMatura) often contain interactive quizzes and exercises that immediately assess the user's answers. In this way, the learner knows immediately whether they have solved the task correctly and receives an explanation for the error. Regular self-checks with generative AI quizzes in Astra AI have proven to be a great tool for monitoring progress - you can quickly identify areas where you need to work harder and go deeper. It was this real-time analysis of my progress that motivated me to use the Math UI app, as it allowed me to track my progress and achievements in real time

An extremely important and not to be neglected feature of AI tutors is their unlimited patience and objectivity. While in the classroom there is sometimes no time to repeat the material over and over again, with an AI tutor or app, a student can repeat the same explanation or task as many times as they want without disrupting, offending or causing outrage by asking a question. You can even ask questions that you would not dare ask in front of the class, for example, let alone in person to the teacher. You can ask the same or a similar question several times and the answer is almost always correct and to the point. With these tools, you are a person with whom the AI has no special or emotional connection, so you have the confidence to ask honestly and without reservation. Online learning platforms such as Astra AI are designed to encourage learning from mistakes - a wrong answer is not an obstacle, but a starting point for further practise, as the AI works with the student until they have understood the material. The student can repeat and correct the tasks at their own pace, building understanding and confidence.

Affordability is also not to be neglected. Most AI learning tools (even in the basic functions) are free or much cheaper than "human" trainers. So, a student can get a personal math "tutor" virtually for free or for a fraction of the price of traditional tutoring. This opens up the possibility of additional learning support even for those who could not otherwise afford expensive tuition.

Observed Limitations and Ethical Considerations

Despite the promising benefits, the use of AI in math education also brings with it a number of challenges and limitations that we need to be aware of. Firstly, there is the issue of reliability. Generative AI tools such as ChatGPT or similar models can sometimes give demonstrably and verifiably wrong or illogical answers. Critically checking the results is therefore an important current limitation and a shortcoming that will be a challenge for all future AI developers. AI tools have proven to me that they can also be unreliable and sometimes even a lazy helper. The problem I see is that AI often sounds very convincing - it formulates answers in a fluent and confident tone, even if they are not correct. If we blindly trust it, it misleads us, and in math that can be a problem. I quickly realised that not all the information the AI gives is always correct. I kept coming across misinterpretations of a problem, incorrect use of full stops and commas, or solutions that were formally correct but did not conform to the methods required by the school system. In fact, Astra AI sometimes uses shortcuts, non-standardised ways that I am not familiar with as a student or that I should not have used in the test. Through this experience I have learnt to check every explanation, sometimes even the same answer several times. I use a combination of consistent, legible and organised notes in class, solving math problems independently in a workbook and reviewing the material using AI tools to learn math. I believe and agree with the renowned experts that without regular active writing in a notebook (Slovenian News 2025) - writing down the material, thinking about the material and solving math problems - the currently developed AI as such cannot replace the classical way of learning and that you cannot achieve the required learning results with it alone. In secondary school, I realised that I do not learn mathematical material thoroughly enough if I accept the explanation too quickly, if I do not understand it and if I do not check its meaning and purpose. If the learning process happens too quickly, I am not able to master the material, which can prove difficult for me to build on. I find the problems my classmates solve in front of the board helpful in class because they expertly guide me in solving the math problems and explain the process and steps required to solve them.

The disadvantage of the math learning AI app under investigation here is that it does not have an inbuilt "push button" or a system of mandatory regular repetition, as is the case with human teacher-led lessons. I believe that if the user consents to the app, the developer should ensure, similar to parental controls on mobile devices, that all functions other than phone and text are disabled if the user has not completed, or completed poorly, the reinforcement tasks assigned to them at a certain date and time. In this case, perhaps a message could also be sent to the user's confidant to encourage them and help them achieve their goal. Motivation to learn is achieved because it is created in the classroom by the educator - the teacher - or because you are forced and you pay a lot or your parents to do so by the hired teacher, whereas AI tools do not currently fulfil this.

Future Potentials and Recommendations

As technology evolves, AI tutoring could go beyond traditional subjects to include soft skills, hands-on practise and interdisciplinary instruction. Emerging technologies such as augmented reality (AR) and virtual reality (VR) are already being along AI also integrated into professional education environments. In Slovenia, such tools are not yet widely used in secondary education, but the groundwork has been laid through numerous Erasmus+ Programmes projects in which many Slovenian vocational school centres are or have been involved and whose usability, applicability and sustainable applications have been tested. In order to fully utilise the educational potential of AI, integration into national curricula, teacher training and the legal framework is essential.

Schools and educational institutions must begin to recognise and promote the sensible use of AI tools. Some Slovenian schools are already organising workshops for students on the use of AI in learning, such as the "Razturi na maturi" initiative on social media, which is funded by the Digital Slovenia (GZS 2024) project. Such initiatives can help to spread digital literacy — so that young people learn to use new tools critically and effectively. However, teacher training also needs to be brought up to date so that teachers can keep up with new technologies and know how to integrate them into math lessons in a meaningful way. The learning tools of the future, supported by AI, should be an extension of the teacher's work, not a competing element. To achieve this, the school system needs to provide guidance, guidelines and the technical infrastructure: from access to devices and the internet to recommendations on how to integrate the use of AI tutors into the curriculum, taking into account the developmental level of students. The Slovenian university of Ljubljana has already gone one step further and has created AI guidelines for studying (University of Ljubljana 2024).

Conclusion

My personal experiences with the use of AI in math lessons are positive, but also complex. As a year 4 (final year) student at secondary school, Astra AI has not only helped me to improve my understanding of more complex math concepts, but also to develop a more structured, independent and responsible approach to learning. Artificial intelligence allows me to learn anytime, anywhere, gives me instant feedback, provides multiple types of explanations and allows repetition without limitations. However, it has challenged me on a few occasions with inaccuracies, misunderstandings and skipped steps when solving tasks. This experience has reinforced my belief that AI is not infallible. I have learnt to check the accuracy of results repeatedly and more thoroughly, to recognise the shortcomings of algorithms and to reflect on the ethical aspects of using such technologies in a school environment. I am convinced that the future of education is hybrid - the human teacher will still play a key role, but supported by advanced technology. However, if in the future we will have for some reason limited or no access to electricity in education, only the trained and experienced educators or andragogs will be able to teach us properly. Therefore, even after the advent of AI, we need to protect and promote human teachers even more.

Astra AI is a compelling example of how targeted, localised AI applications can improve student learning, especially in challenging areas such as math. In my personal experience, it has not only improved my understanding and confidence, but also provided insights into the evolving role of digital tools in education. As AI becomes more ubiquitous, its pedagogical role needs to be carefully chosen to maximise the benefits and minimise the risks. Hybrid education— - a blend of human tutoring and AI support — promises a more accessible, efficient and student-centred future.

Sources:

Astra AI offers free access to the first AI math tutor to all teachers in Slovenia (2024, 18.3.). Svet24. Available at: https://svet24.si/magazin/njena/clanek/astra-ai-matematika-ucitelji-slovenija-1382599 (retrieved 28.2.2025).

About the project - Course materials (2024, February). Digital Bridge. Available at: https://www.digi-most.si/o-projektu (retrieved 28.1.2025).

AI as a tutor - How to learn effectively with AI (2024, March). eMatura. Available at: https://www.ematura.si/2024/10/04/ai-tutor/ (retrieved 12 February 2025).

ChatGPT 3.5 - Introduction to ChatGPT. (2024, February). Digital Bridge. Available at: https://www.digi-most.si/programi/chatgpt-3-5---uvod-v-chatgpt (retrieved 27 January 2025).

GZS.SI-Chamber of commerce and industry of Slovenia (2025, March) The importance of teaching materials in building a better future. Available at: Pomembnost učnih gradiv pri gradnji boljše prihodnosti - Media GZS (retrieved 22 March 2025).

If you want to start using AI tools, put them together with your children and teach them (2024, April). RTV Slovenia. Available at: https://www.rtvslo.si/znanost-in-tehnologija/ce-zelite-zaceti-uporabljati-orodja-umetne-inteligence-jih-dajte-skupaj-z-otroki-in-jih-naucite/735876 (retrieved 15 February 2025).

Kovač Polajnko, T., Jereb, E. and Cugelj, L. (2024) Artificial intelligence in education: comparing the responses of different large-scale language models. Contemporary Pedagogy, 75, no. 1, pp. 113-130. Available at: https://www.sodobna-pedagogika.net/clanki/04-2024_umetna-inteligenca-v-izobrazevanju-primerjava-odzivov-razlicnih-velikih-jezikovnih-modelov/ (retrieved 28 February 2025).

The Free ChatGPT Prompt Generator. (2024, March). Feedough. Available at: https://www.feedough.com/chatgpt-prompt-generator/ (retrieved 26 February 2025).

The AI Factor - How does artificial intelligence mimic the brain? (2024, April). RTV Slovenia. Available at: https://365.rtvslo.si/arhiv/ui-faktor/175120252 (retrieved 23 March 2025).

Siol.net. (2024, March). Astra AI - A math app already used by more than 70,000 Slovenian students. Available at: https://siol.net/novice/slovenija/astra-ai-aplikacija-za-matematiko-ki-jo-uporablja-ze-vec-kot-70-tisoc-slovenskih-ucencev-656687 (retrieved 10 March 2025).

University of Ljubljana. (2024, April). The use of artificial intelligence in education: methodology and state of the art review. Available at: https://www.uni-lj.si/assets/Center-Digitalna-UL/Pristopi/Smernice-UI-v-izobrazevanju/Porocilo-UI-v-izobrazevanju-pregled-stanja-in-metodologija.pdf (retrieved 28 February 2025).

Writing for the brain: handwriting improves memory. (2025, March). Slovenian News. Available at: https://www.slovenskenovice.si/ona/aktivni-in-zdravi/pisanje-za-mozgane-pisanje-na-roko-izboljsuje-spomin/ (retrieved 27 March 2025).