How Alef uses AI for Personalised Learning

Diana (Fangyuan) Yin (she/her/hers)
Alef Education
Published in
8 min readFeb 17, 2020


In the past decade, “personalized learning” has become a buzzword in education. In traditional classrooms, students are taught in the same way, using the same materials, and at the same speed. After big data and AI stepped in, the one-size-fits-all model doesn’t have to be the only option anymore. AI has started a brand-new trend that is transforming educational practice.

Personalized learning has become a buzzword in education in recent years. Photo by Headway on Unsplash

How does AI provide a personalized educational solution?

To give a convincing answer to this question, in this article, I will look into my current company, Alef Education, to unveil its six AI solutions. Alef Education is an Ed-Tech company based in Abu Dhabi, UAE. The mission of Alef is to transform K-12 school systems with technology-enabled, multimedia-enhanced learning experiences that engage a more individualized sense of inquiry and empower the 21st-century workforce. From 2015, the year that the idea of Alef first came about, until now, Alef Education has become the industry leader of personalized learning in the Middle East and beyond.

AIef’s six AI solutions include

  1. Student knowledge tracing system,
  2. Remediation recommendation system,
  3. Evaluation of assessment questions engine,
  4. Final exam prediction model,
  5. Automatic short answer grading, and
  6. Adaptive diagnostic test.

From an educational point of view, these six solutions fall under three aspects, which are three steps in the process of personalized learning: 1) personalized knowledge acquisition, 2) personalized assessment, and 3) personalized prevention and intervention.

Personalized knowledge acquisition

This educational endeavor is made possible via two channels: student knowledge tracing system and recommendation engine built on top of the knowledge tracing system.

Imagine a learning assistant that detects how much knowledge each student has mastered in real-time, as they move through their learning materials, and accordingly makes recommendations when a weak knowledge point is noticed. Sounds exciting, right? Alef is well on its way to building such a learning engine.

The knowledge tracing and recommendation systems are achieved by considering several elements, among them, subject-specific Knowledge Graph, and student interaction with the product.

Starting with Math and Science, Alef Education utilizes state-of-the-art AI technologies to break the task of learning into bite-sized learning objectives. Subject skills are extracted from unstructured data, such as lessons and books, and internationally adopted curriculum standards, and automatically tagged with the lessons and learning outcomes. This process leads to building a structured knowledge base consisting of skills, lessons, and prerequisite relationships. The relationship is represented as a network-graph called Knowledge Graph. Educational experts evaluate and assist AI to improve the quality of the extracted skills and knowledge graph.

Alef Education is using AI to transform how children learn. Picture from

Using the Knowledge Graph, Alef efficiently determines and traces student’s skill-mastery for each subject, by also considering individual student’s interaction with the Alef platform. With every single click, every second spent on each page, and every question answered or failed, students keep “telling” the AI engine how much they have mastered, where the gaps are, how they prefer to learn, what they perceive to be difficult or easy.

With a myriad of data, Alef not only determines whether a student has mastered a skill and traces the mastery evolution over time but also establishes a recommendation system that suggests to students the practices and tasks to improve their low-performing skills.

With the power of big data and AI, Alef Education constructs a knowledge acquisition approach where each student is learning in a hyper-personalized way.

Personalized assessment

Alef Education’s solution in personalized assessment comes in two aspects: 1) evaluation of assessment questions, and 2) automatic short answer grading.

1) Evaluation of assessment questions

One of the common mistakes in the industry might be that educational companies are not paying enough attention to students’ involvement with and feedback on assessment questions. Content creators are busy writing assessment questions, while the opinions and input of students are rarely taken into consideration.

It takes more effort to generate questions than answer them. Photo by João Silas on Unsplash

Alef Education resolves this issue in this way: as usual, students finish the questions and submit their answers. Based on that, and with the solution of Item Response Theory (IRT), the system then evaluates the difficulty level and quality of assessments to understand the design of question items. In this way, Alef students are continuously “telling” question designers how they think of the questions. The results could help Alef achieve a balance of “easy,” “medium,” and “difficult” question items, and evaluate and improve question design.

For example, through existing analysis, Alef AI shows that 1) the difficulty level of Alef assessment questions follows a normal distribution (the bell curve), with the majority of questions falling under “medium” difficulty level, and smaller number of questions labeled as either “easy” or “difficult”; 2) Compared with numerical Math problems, Alef students struggle with word problems, which might be explained by their struggling reading skills. On top of that, Alef AI improves assessment in many ways with the same results, including in-house adaptive testing, assessment normalizing, faster grading, and tracking steps in problem-solving.

2) Automatic short answer grading

Multiple-choice questions are the most popular assessment tool. Source: Shutterstock

Automatic short answer grading is the technology where short, human-generated, natural language responses to objective questions are assessed using computational methods. Let’s face it: compared with short answer questions, multiple-choice questions (MCQ) have multiple disadvantages. Students can more easily guess correct answers with luck. They are relatively easier since they don’t require students to generate anything. And it’s harder for MCQ tests to assess higher-order cognitions, such as critical thinking, synthesis, and evaluation.

For these reasons, Alef Education is introducing automatic grading of short questions. Students get to type their short responses to the platform, and the AI engine uses the NLP techniques to “read” their answers and provide grades immediately, ridding students of the long wait to receive feedback on their responses, and freeing assessors from the pain of grading student answers.

Personalized prevention and intervention

In schools, early-warning and intervention programs have been in place for a long time to help learners who are on the edge of failing their exams and dropping out of schools. Research has proven the effectiveness of preventative approaches in engaging high schoolers, identifying potential dropouts, and improving school graduate rates [1] [2].

In the context of the e-learning environment, early-warning data systems can equip schools with more intact solutions to help struggling students and provide preventative strategies. In the context of Alef Education, a personalized preventative and intervening mechanism is built by combining 1) final exam prediction and 2) adaptive diagnostic tests.

Reducing the dropout rate remains one of the main educational targets. Source: Shutterstock

1) Final exam prediction model

Based on the system’s understanding of students, Alef predicts each student’s probability of failing in the upcoming final exam and classifies students into different intervention groups accordingly. The system identifies not only poor-performing learners but also exceptionally talented ones, answering their personalized requirement for guidance.

The parameters that contribute to the prediction model include, but are not limited to, student responses to Alef platform questions, performance on previous assessments, the number of days that students are active on the Alef platform, and the number of completed Alef lessons. A combination of these (and more) parameters are closely tied to educational factors, such as student dedication to learning, engagement, baseline proficiency level, etc.

In this way, Alef formulates a continuous tracking system that signals abnormalities at an early stage so that Alef employees and school teachers could introduce interventions pre-emptively.

2) Adaptive diagnostic test

As early as 2011, the UAE Ministry of Education has launched a diagnosis to public schools to assess student academic levels. The role of diagnostic tests helps prevent pre-existing assumptions about student language proficiency, actual grade level, and foundational skills. Such an effort re-adjusts the expectation of teachers and curriculum designers and assists them in identifying problems early on.

Photo by Green Chameleon on Unsplash

Accordingly, Alef has embarked on the integration of diagnostic tests as part of our product. At the beginning of each academic year, both existing and new Alef students will come back to their study and receive a diagnostic test for English-medium subjects (English, Math, and Science). The test would inform both Alef team and school teachers of student level, and how much they could potentially understand their current grade level’s materials. When the test starts, the system would begin with a random level and based on the student response, move to a higher or lower difficulty level. At the end of the test, the system would detect multiple areas of expertise in that subject, and span over several grades to give a thorough presentation of student skills.

Behind the scene, Alef AI technology and Alef in-house teams power this endeavor. From the technology point of view, Item Response Theory (see details in “Evaluation of assessment questions”) determines the difficulty level of questions, and the next question to present right after students submit the previous one. Besides, human-led, machine-assisted content and question tagging mechanisms make sure that every piece of question from the question pool is mapped to larger-scale learning objectives, so student performance on the diagnostic test isn’t independent of their overall learning plan.

To summarize…

With the power of AI, Alef aims to establish a holistic educational experience, from the moment students acquire new knowledge, to the step when the skills are reviewed and evaluated. The solutions Alef presents also target both high-performing students and at-risk students, leaving no student behind and catering to personalized needs.

What do you think of Alef AI solutions? Do you know other companies that offer similar approaches? Share your opinions with us by commenting below.



Diana (Fangyuan) Yin (she/her/hers)
Alef Education

Product Manager. Harvard GSE. Michigan Ross MBA Candidate. CFA. In tech industry for 6 years. I write about tech for fun. Writing to fulfill my childhood dream.