VNPT partners with the Posts and Telecommunications Institute of Technology to build an AI-native university model
This partnership is part of VNPT's overarching strategy to develop an AI Innovation Hub, leveraging synergies with prestigious domestic and international institutes and universities to address real-world challenges in Vietnam.
On September 5, in the presence of Mr. Vu Hai Quan—a member of the 14th Party Central Committee and Minister of Science and Technology—the Vietnam Posts and Telecommunications Group (VNPT) and the Posts and Telecommunications Institute of Technology (PTIT) signed a cooperation agreement to develop artificial intelligence for education.
The partnership focuses on researching and mastering specialized Vietnamese Large Language Models (LLMs) for education, developing an AI Tutor Platform, and gradually building a comprehensive AI-integrated university model—an "AI-Native University."
AI in education: not just "knowing," but being "correct"
Education sets exceptionally high standards for artificial intelligence. An AI tutor cannot simply generate answers that sound plausible; the system must be factually and logically accurate, closely aligned with the curriculum, tailored to the learner's proficiency, and capable of explaining why a solution is correct or incorrect.
Therefore, the challenge of AI in education goes beyond merely building a model that can answer a wide range of questions. It is crucial to control where the AI sources its knowledge, how it reasons, and whether the final output is reliable enough to be presented to learners.
According to a VNPT representative, VNPT AI develops Vietnamese LLMs of various scales to meet specific deployment requirements, with the largest model boasting hundreds of billions of parameters. To date, the system has processed over 1.8 trillion tokens, addressing real-world challenges across sectors such as finance and banking, insurance, telecommunications, the public sector, and others.
VNPT AI focuses not only on model scale but also on factors critical to practical AI implementation, including reasoning capabilities, output quality, and operational efficiency. A key challenge for these models is enabling the system not only to provide an answer but also to pinpoint exactly where a reasoning process goes wrong.
VNPT AI has developed a method allowing Large Language Models (LLMs) to automatically verify mathematical reasoning chains, identify the specific step where an error occurs, and classify the cause of the mistake. This method was presented at AI4Math 2026—held in conjunction with ICML 2026 (a top-tier A* conference on machine learning, deep learning, and modern AI methods)—where it ranked in the top two.
This capability is particularly significant for AI tutoring applications. Instead of simply telling a student their answer is incorrect, an AI tutor can emulate the approach of a human teacher by identifying the specific step where the student erred and helping them understand the underlying cause of the mistake.
On a broader scale—such as within a school environment—a single subject may involve hundreds of different curricula, reference materials, lectures, and knowledge sources. If a system merely searches for and aggregates information, the results may suffer from omissions, redundancies, inconsistencies, or even contradictory content across sources.
To address this, VNPT AI developed a method combining multiple AI agents with a knowledge graph, enabling the system to read and synthesize information from various documents simultaneously rather than processing sources in isolation. The technology's value lies in its ability to cross-reference multiple sources, minimize information gaps, reduce redundancy, and detect potentially contradictory content.
This method was also published in the Q1 international journal Neural Computing & Applications (2026). When integrated into an AI tutor, this capability allows the system to leverage vast amounts of curricula and learning materials while maintaining a unified, evidence-based, and verifiable knowledge base.
 |
From AI Tutors to the AI-Native University
Under the partnership agreement, the scope of collaboration between VNPT and PTIT extends beyond AI tutors. The two parties aim to build a comprehensive AI-integrated university model—an "AI-Native University"—where AI supports a wide range of activities, from education, research, and assessment to administration and student services.
This partnership, involving the State, the university, and the enterprise, exemplifies the "Triple Helix" model by connecting state strategic direction, the university's research and educational capabilities, and the enterprise's technological and implementation strengths to apply AI to real-world educational and societal challenges.
While an AI tutor addresses a specific touchpoint in the learning process, the AI-Native University model presents a broader scope. In this model, faculty gain tools to design and personalize educational activities; students have an on-demand learning assistant; researchers can more effectively leverage knowledge and data; and the university can utilize AI to support administration, decision-making, and service delivery.
PTIT serves as the environment for research, experimentation, and practical implementation, where AI capabilities are directly validated against higher education challenges, leveraging its strengths in faculty, experts, academic data, pedagogical standards, and its technology education ecosystem.
Conversely, VNPT leverages its proven core AI technologies, computing infrastructure, and experience in deploying large-scale digital platforms to research, develop, and gradually operationalize these AI models.
