Building a Data Management System and AI Assistant for Quality Assurance at Tay Nguyen University

Building a Data Management System and AI Assistant for Quality Assurance at Tay Nguyen University

1. Context and Problem statement

Tay Nguyen University (TNU) is a MOET-affiliated public university with a clear governance model (Party Committee, University Council, Board of Rectors), 9 functional offices, 9 academic units, 2 training-support units, and 3 practice/teaching facilities. As of June 2025, TNU had 635 staff members (of whom 349 female and 286 male; 30 are from ethnic minority groups), of whom more than 60% are academic staff (01 Professors, 16 Associate Professors, 102 PhDs, 04 Specialist Level II, 06 Specialist Level I physicians, 258 Masters, 69 Bachelor’s.

Tay Nguyen University (TNU) is striving to enhance its educational quality and achieve rigorous accreditation targets for 2020–2030. The university has established a dedicated Quality Assurance Office with 10 staff members to lead internal quality assurance (QA) efforts. TNU has been actively pursuing both institutional and program-level self-assessments, and in 2022 it became an Associate Member of the ASEAN University Network – Quality Assurance (AUN-QA). By late 2024, two academic programs (Economics and English) had already been accredited under AUN-QA standards, marking initial successes in external quality verification. The strategic goal is that by 2030, 100% of TNU’s programs achieve domestic accreditation and at least 20% attain international accreditation, aligning with national higher education quality benchmarks. This push comes amid strong directives from the government and university leadership to accelerate digital transformation and apply artificial intelligence (AI) in governance and quality management. In March 2025, the Prime Minister approved a plan for TNU (and a peer institution) to become regional centers of excellence, which explicitly calls for “boosting digital transformation and AI applications in teaching and university management”.

Despite this supportive policy environment, TNU’s internal QA system faces significant challenges. Quality-related data are currently siloed across multiple platforms and formats – for example, some data reside in the national Higher Education Management Information System (HEMIS) and the MOET’s SAHEP project system, while many accreditation evidences are stored in disparate databases or even in manual files. There is no centralized QA data repository to enable efficient access and analysis. (Notably, the Ministry of Education and Training has deployed HEMIS nationally to collect data from all universities, but at the institutional level TNU lacks an integrated database for QA evidence, causing fragmentation.) Self-assessment reports – required for accreditation – are still prepared largely by hand. QA staff must gather evidence and statistics from various sources and manually draft narratives for each criterion, which is labor-intensive and time consuming. This manual process often leads to duplicated information across reports and may fail to fully leverage the abundant evidence available. Although the university has an existing software system for managing accreditation evidences and aiding report writing, it currently has minimal AI capabilities. In practice, report writing teams receive little intelligent support from the system beyond basic document storage, and thus spend upwards of 3 months assembling a self-evaluation for one program. This is inefficient and risks inconsistencies or omissions in the final report.

Furthermore, stakeholder feedback analysis is underdeveloped. TNU’s QA office conducts regular surveys to collect opinions from students and other stakeholders on teaching quality, courses, facilities, etc. (for instance, end-of-semester teaching evaluations, graduate satisfaction surveys, etc., are administered each year). However, analyzing these feedback data is done manually and mostly limited to basic statistics (e.g. average ratings, simple charts). Such traditional analysis is often slow and superficial – it may miss deeper insights hidden in open-ended comments or trends across semesters. Studies note that manual feedback analysis is not only time-consuming but can be subjective and fails to promptly yield detailed actionable information from large volumes of responses. In TNU’s case, open text responses from students (which could contain sentiments and recurring issues) are not systematically analyzed for sentiment or themes, meaning potential quality problems or improvement opportunities might remain latent.

Most critically, the university lacks an early warning system for student performance risks. There is currently no data-driven mechanism to proactively identify and support students who are struggling academically or at risk of dropping out. Interventions usually rely on lagging indicators (like end-of-semester GPA or eventual academic probation), by which time it may be too late to effectively retain the student. Research shows that many universities’ warning and counseling efforts come “too late” – typically after final grades or when a student has already initiated withdrawal – and thus have limited effect on reducing attrition. Proactive models using machine learning can predict student dropout risk much earlier by analyzing various indicators (grades, attendance, library use, advising sessions, etc.), enabling timely support and reducing the dropout rate. TNU currently has no such predictive analytics in place; any support to at-risk students is reactive. This gap is especially pressing as national standards now consider a high dropout rate (e.g. >10%) as a serious quality issue for a university.

In summary, TNU’s internal QA system is not yet keeping pace with its growth and the increasing demands of accreditation and digital governance. Data scattered in multiple places, labor-intensive report writing, rudimentary feedback analysis, and lack of early warning tools all hinder the university’s ability to assure and enhance quality efficiently. These limitations persist even as TNU faces pressure to innovate: the government and university leaders expect robust digital QA systems and AI-driven management as part of the digital transformation agenda. This contrast between external expectations and internal capability constitutes the core problem. The urgent need is to integrate AI technologies into the QA system to manage data more effectively, automate analysis and reporting, and ultimately improve educational quality in a sustainable way.

Key challenges summary: TNU’s IQA system lags behind accreditation and digital governance demands. Data dispersion, labour‑intensive reporting, rudimentary feedback analytics, and lack of predictive support hinder QA effectiveness. Meanwhile, policy expectations require robust, AI‑enabled QA. The urgent need is to integrate AI into QA to manage data, automate analysis/reporting, and improve quality sustainably.

2. Project Objectives

Overall Goal: The project aims to significantly improve the effectiveness and automation of TNU’s internal quality assurance system through AI applications. By leveraging AI for data integration, analysis, and report generation, the university will streamline QA processes (data collection, self-assessment reporting, feedback analysis) and better meet its accreditation and digital transformation targets. This contributes to sustaining high educational quality and supporting student success, in line with TNU’s strategic plan and accreditation commitments.

Specific Objectives and Key Performance Indicators (KPIs):

• Centralized QA Data Management: Develop a unified, central database that consolidates all key quality assurance data – including accreditation evidences, survey results, student learning data from HEMIS, outcomes (graduation rates, employment), etc. KPIs: By December 2025, have an integrated QA data repository operational, containing at least 80% of the important QA indicators and evidences needed for self-assessment. Success will be measured by data coverage and reduction in time spent by staff to retrieve information. Rationale: A unified database addresses the current fragmentation – when all QA evidence is in one place, it becomes feasible to apply analytics and AI.

• AI Assistant for Self-Assessment Reporting: Implement an AI-powered module (using Natural Language Processing) to support the writing of accreditation self-evaluation reports. This “QA AI Assistant” will suggest content, summarize evidence, and draft sections aligned to accreditation standards. KPIs: Reduce the time to prepare a self-assessment report by 30–50% (≈3 months → ~1.5 months). By 2026, AI-generated drafts/suggestions used for ≥50% of criteria. Rationale: Automating retrieval and narrative drafts improves coherence; recent QA forums report significant time savings.

• Enhanced Feedback Analysis and Early Warning: Utilize AI to analyze stakeholder feedback (sentiment, themes) and pilot a predictive model to flag at-risk students. KPIs: From 2026 (after April 2026, mainly for Phase 2), produce ≥2 AI-driven feedback analysis reports/year; achieve ≥70% correct identification of at-risk students in tests. Rationale: NLP unlocks open-text insights; ML early-warning reduces attrition.

• Build Digital QA Capacity: Train QA staff and relevant faculty to use the system and embed new AI-enabled QA practices. KPIs: Train 100% of QA Department staff and ≥30 additional staff by 2026; develop ≥2 official AI-in-QA workflows (AI-assisted reporting; advisor protocol for alerts).

Change priorities:

(1) Curriculum relevance via data-driven QA (by Jan 2026).

(2) Industry partnerships via Centre for Innovation and Entrepreneurship/Advisory Committees (1–2 mechanisms by Dec 2025).

(3) Regional/international integration & benchmarking (beyond 2026).

3. Scope of Implementation

This project will be implemented university-wide, with an initial focus on key pilot areas to ensure effectiveness before scaling up. Pilot Units: one priority program preparing for AUN-QA accreditation (AI-assisted report writing) and the QA office for institutional self-assessment and dashboards. Data Coverage: integrate data from Training, Student Affairs, QA, faculties, and HEMIS/ministerial systems to build a multidimensional dataset. Stakeholders and Beneficiaries: QA staff and program teams (primary users), leadership (dashboards for decisions), advisors/mentors (early alerts); students benefit indirectly. Scaling Plan: after pilot success, expand to all faculties/programs with modular architecture, phased onboarding, and continuous model/data updates.

4. Main Activities and Timeline (P-D-C-A)

Phase 1 – Plan and Do (November 2025 – January 2026)

– Detailed Needs Assessment and System Design (November 2025 – January 2026): Map QA data sources; interview stakeholders; define requirements; assess build vs. extend options; draft database schema and AI assistant functions (evidence summary, draft per criterion); plan feedback analytics and early-warning models.

– System Development and AI Module Training (January 2026 – March 2026): Implement unified QA database and hosting; migrate initial data; fine-tune NLP assistant on past reports/evidence; implement sentiment pipeline; prototype risk model (e.g., logistic/tree-based) per research; design QA dashboard UI.

– Pilot Deployment and Refinement (March and April, 2026): Use system in a real self-assessment (Run one live SAR with AI); run AI-based survey analysis; test early-warning on recent cohorts; gather metrics vs KPIs; iterate models/UI; train power users.

Phase 2 – Check and Act (May – December 2026 and beyond)

– Evaluation of Pilot Outcomes (May – September 2026): Measure KPIs (data coverage, time reduction, AI usage, prediction accuracy, user satisfaction); external QA review of report quality; document lessons and adjustments.

– Scaling Up and Institutionalization (Post-2026 – Act): Phased rollout to all programs; policies mandating system use; assign QA/IT ownership; role-based access, privacy and backups; continuous model retraining and tech upgrades.

Monitoring and completion :

·       KPI set: Increase the number of surveys and the associated response rates; conduct at least one Focus Group Discussion or seminar for each program; workshops & participation +20%; ≥1 advisory committee/faculty; 40 programs revised; MoUs 50→100/3y; international students 30→90–100/3y; mobility 10–20/AY; employability 80%→90%; ranking 73→60/3y; enrolment +10% (2,200→≥2,500/3y); improved entry quality in ≥10 majors.

·       Completion: KPIs met/near-met; stakeholder satisfaction evidenced; internal/external benchmarking confirms progress.

5. Expected Results/Outcomes

1. Integrated QA Data and Analytics Platform: A centralized database with real-time dashboards (graduation rates, employment, student satisfaction, etc.), improving transparency, reducing duplication, and enabling leadership to access a “QA health snapshot” anytime.

2. AI Assistant for Self-Assessment Reports: An NLP-powered module to draft accreditation report sections, link evidences, and suggest strengths/weaknesses. Expected to cut report preparation time by ~50% and improve completeness and consistency, leading to stronger accreditation outcomes.

3. AI-Enhanced Feedback Analysis and Early Warning: At least two AI-generated feedback analysis reports per year, revealing sentiment and themes from student comments. A predictive model will flag ≥70% of at-risk students, enabling proactive advising and reduced dropout rates.

4. Improved QA Efficiency and Culture: QA processes become faster and more data-driven; staff shift from clerical tasks to improvement actions. Dashboards and alerts reinforce continuous improvement and shift mindsets from compliance to development.

5. Strengthened Digital and AI Skills: ≥40 QA staff and faculty trained in AI-enabled QA practices. New workflows institutionalize AI use, leaving a sustainable legacy of a tech-savvy QA workforce.

6. Risk Management

Implementing an AI-based QA system is ambitious, with several risks and barriers. Mitigation strategies are as follows:

1. Technical expertise and data quality: TNU has limited AI/data expertise and fragmented QA data. Mitigation: Engage external experts and mentors, train QA staff through workshops, and adopt a phased integration—prioritizing key data first, while cleaning and standardizing over time.

2. Financial and infrastructure constraints: Developing AI tools requires funding and computing resources. Mitigation: Maximize existing infrastructure, use open-source frameworks and affordable cloud services, and seek additional budget support or external resources (e.g. InnoAIQA toolkit). Start with scalable, low-cost models and expand gradually.

3. Human/Staff resistance to change: Staff may be skeptical of AI or reluctant to change workflows. Mitigation: Ensure leadership mandate, demonstrate quick wins (e.g. AI-generated drafts saving time), use early adopters as role models, provide hands-on training, and emphasize AI as a support tool. Build trust by validating AI outputs during pilots.

4. Sustainability and maintenance: Risk of system disuse without clear ownership; security and privacy also critical. Mitigation: Assign QA office as system owner, IT Center for technical support. Establish maintenance schedules, train multiple staff, integrate use of the system into official QA procedures, and enforce strict data security (role-based access, anonymization, backups). Continuous user feedback and updates will keep the system relevant.

7. Support Requested from the InnoAIQA Program

Technical consultation and AI expertise. Assign experts in Vietnamese NLP and education analytics to advise our report-writing assistant and student-risk model (e.g., scoping, data preparation, algorithm/feature selection, evaluation). Provide periodic clinics/code reviews, reference case studies, and any InnoAIQA toolkits/libraries.

Mentoring and peer learning. Pair us with a project mentor experienced in HE QA/data systems for milestone reviews and risk management. Facilitate peer exchanges (virtual or onsite) with institutions running digital QA dashboards, AI-enabled feedback analysis, or early-warning systems.

Tools, templates and resources. Share ready-to-use dashboard and survey templates; data-governance guides; sample QA data schemas; sentiment/NLP starter code. Where possible, offer time-limited credits or licenses to evaluate cloud AI services (e.g., Azure/Google NLP) to reduce prototyping costs.

Networking and future collaboration. Connect us with International QA and national QA networks for benchmarking and good practices. Help disseminate project outputs (workshops/conferences), and provide endorsements that strengthen bids for follow-on funding and partnerships.