Universitas Terbuka enhances distance learning for 100,000 students with Azure AI

Indonesia’s Universitas Terbuka (UT), with over 525,000 active students and growing, struggled to provide timely tutor support, risking student engagement, satisfaction, and retention.

UT used Microsoft Azure OpenAI Services and Azure AI Foundry to build an AI tutor that delivers accurate, curriculum-aligned responses and streamlines student assessment. The tutor currently supports 500 classes and some 100,000 students.

The AI tutor enhances student grades and engagement by providing more comprehensive support 24 hours a day. It also reduces educators’ workloads and frees them to focus on academic excellence while enabling the university to continue growing.

Universitas Terbuka (UT), Indonesia’s largest public university by enrollment, makes education accessible to all Indonesians, regardless of their location anywhere in the world. In recent years, the convenience and low cost of its open distance model, plus growing student comfort with remote learning, have driven exploding enrollment. More than 525,000 active students—a number projected to reach 750,000 before 2025—have strained the university’s resources.

Providing timely, effective, and personalized tutor support has been particularly difficult on such a large scale. Many of the university’s educators are responsible for hundreds of students in up to a dozen different classes. Each semester, UT sets up online tutorials for each, creating tens of thousands of virtual tutorial classes that involve up to 2.3 million student interactions in total over the term. It could take instructors as long as a week to reply to student questions or assignments submitted through UT’s online learning management system (LMS). The challenge was especially frustrating and time-consuming for educators tasked with repeatedly answering similar questions. This situation affected student engagement, progression toward degrees, and ultimately student retention. To improve student satisfaction, achieve its mission, and continue to grow, UT needed a new approach.

Supporting educators and students with AI

To continue providing quality education without overburdening educators, Universitas Terbuka turned to generative AI and Microsoft Azure AI, data, and app services solutions. UT uses a hybrid application infrastructure, shifting workloads between on-premises hosting for inactive courses and a microservices architecture on Azure. Cloud hosting delivers the scale needed for active courses with peaks of up to 60,000 concurrent users. That ability to scale made Azure AI the obvious choice for a university anticipating 1 million active students before 2027.

"We needed a solution that is scalable and stable enough to handle the number of students. Azure OpenAI Service provided that scalability,” says Dimas Agung Prasetyo, M.S., currently Head of the Undergraduate Program in Data Science for Universitas Terbuka. Previously Director of Information Systems, he initiated the AI project to enhance learning services for students.

We needed a solution that is scalable and stable enough to handle the number of students. Azure OpenAI Service provided that scalability.

Dimas Agung Prasetyo, M.S., Head of the Undergraduate Program in Data Science, Universitas Terbuka

In 2023, the university used Azure AI Foundry to build an AI tutor accessed through the school’s customized Moodle LMS, which students use for discussions and assignments. The AI tutor uses Azure OpenAI Service for the latest GPT series models, and Azure AI Foundry enables parallel testing of each model. Prasetyo says, “We used Azure AI Foundry as our integrated development environment.”

Ensuring accurate response quality

Course lecturer materials and grading rubrics for five core classes, including Indonesian Language and Civic Education, were indexed. Source documents were scanned efficiently with the help of Azure AI Document Intelligence, part of Azure AI services. The results were stored in object storage using Azure Blob Storage or as vectors using Azure Cosmos DB to streamline information management.

We used Azure AI Foundry as our integrated development environment.

Dimas Agung Prasetyo, M.S., Head of the Undergraduate Program in Data Science, Universitas Terbuka

“The indexing went faster than we expected,” notes Prasetyo, who says the resulting knowledge base is also easy to update. Retrieval Augmented Generation (RAG) with Azure Cosmos DB helps ensure accurate tutor responses aligned with instructors’ curricula. Although the AI tutor can access external knowledge bases, its responses rely primarily on the internal knowledge base. Prasetyo says replacing the university’s traditional MongoDB services with Azure Cosmos DB also accelerated development and AI response generation. Finally, Azure Cosmos DB works with the university’s PostgreSQL database that manages general LMS data storage and caching.

This architecture reliably and automatically accommodates high-volume, concurrent interactions with students and the generative AI model. Meanwhile, Azure Kubernetes Service (AKS) and Azure App Service support several key academic services apps, which track students’ class registrations and degree progress.

More about this diagram

“Azure Kubernetes Service lets us deploy in a microservices architecture in the easiest way—automatically and with less pain,” Prasetyo says. “And Azure App Service makes deployments faster, too. We don’t have to do much fine-tuning. We trust that Azure is fine-tuning for us.” UT’s modernized Azure infrastructure also reduces IT workloads by minimizing maintenance of its on-premises datacenters.

Making education more flexible and inclusive

Because students need both answers to questions and feedback on completed assignments, UT also needed the optical character recognition (OCR) capabilities of Azure AI Vision, another of the Azure AI services. This solution enables students to submit text-based work for assessment, either by entering answers online or by uploading Microsoft Word or PDF documents.

The resulting tutor is awake and working nonstop, making education more available, flexible, and inclusive. Students can learn and progress more quickly because they receive quick answers and other support. For instance, the AI tutor directs students to additional information in course reading materials and suggests ways to improve future assignments.

In the initial pilot, the AI tutor answered student questions for about 500 students enrolled in one section each of five core classes. The pilot validated the system architecture, integration, and operational costs. A second phase expanded the AI tutor to some 100,000 students, demonstrating scalability. Prasetyo hopes to double the number of classes involved to 1,000 soon. Ultimately, the AI tutor will aid students in all UT classes.

Azure Kubernetes Service lets us deploy in a microservices architecture in the easiest way—automatically and with less pain. And Azure App Service makes deployments faster, too.

Dimas Agung Prasetyo, M.S., Head of the Undergraduate Program in Data Science, Universitas Terbuka

Improving student grades 24 hours a day

The always-on availability of the AI tutor has made UT education even more flexible and increased student engagement and productivity. Students’ questions receive more comprehensive answers, and they can follow up for interactive learning. This enhances knowledge retention, academic performance, and student retention.

“Students who have the AI tutor in their classes have the ability to get better grades than students who don’t,” says Prasetyo. “That’s good news.” Specifically, research data from four courses involving nearly 38,000 students indicated that those with the AI tutor participated more in discussions and earned better scores on assignments. These results were statistically significant.

The AI tutor has reduced the workload of lecturers, who act as moderators. Prasetyo notes, “We don’t have to answer student posts one at a time. We can just curate the AI responses.” The AI tutor also speeds the creation of assignments from days of work to just one or two. As a result, educators can focus on more impactful activities such as curriculum enhancement, research, and personalized student support.

UT is currently testing an AI tutor enhancement for working with mathematical problems, including the ability to recognize handwritten solutions. This new capability is expected to go live soon. It will enhance the educational experience for many more students and classes.

Prasetyo says, “As our student load is increasing, the AI tutor gives better service to our students by giving more responsive, more elaborate answers while also reducing tutor workloads and optimizing resources to help keep tuition costs low. It’s enabling Universitas Terbuka to continue to grow as an education innovator.”