Developing a Lao-Language Chatbot with RAG to Enhance Education Services : A Case Study at Savannakhet College of Health Sciences

다국어 초록 (Multilingual Abstract)

In this thesis, the design, implementation, and evaluation of a modern website and Lao language AI agent of the Savannakhet College of Health Sciences are described. The project helps overcome shortcomings in the old library resources, lack of coordination of services provided to students, and inability to provide real-time student interaction with the institution. The content management system
(CMS) was created to deliver news, notifications, and downloadable academic resources, whereas the AI chatbot, based on official college materials and professional advice, provides 24/7 answers in the Lao language. Technically, it was constructed with the help of Flutter, DART, Firebase, and Vertex AI, and retrieval-augmented generation (RAG) was utilized to enhance accuracy and decrease hallucination. The test was a combination of automatic BLEU-score benchmarking on 36 questions related to diabetes and surveys of students carried out prior to and following the updates. The findings indicated a small percentage increase in chatbot accuracy (~53-56 percent) and a great percentage increase in student satisfaction especially in clarity, consistency and information accessibility. The findings suggest that AI in Lao-language services may be applied to enhance the transparency, reduce the job of the staff members, and enhance student experiences in higher education. In addition, the research article illuminates on the potential and the danger of operationalizing advanced AI technologies in low-resource economies such as Laos, and suggests future studies, including dataset expansion, superior embeddings, and superior regulation, to attain safety and sustainability.

목차 (Table of Contents)

• 1. Introduction 1
• 1. Problem Statement 1
• 2. Objective & Goals 3
• 3. Impact & Significance 3
• 2. Literature review 3
• 1. Open-Source LLMs: Llama Series 4
• 2. Google DeepMind’s Gemini 5
• 3. LLMs and Healthcare 6
• 3. LLMs and Education 8
• 4. LLMs in the Lao PDR 9
• 5. Natural Language Processing (NLP) 10
• 6. Transformer Models 12
• 7. Encoder–Decoder Architecture in AI 14
• 8. RoRA: Rank-adaptive Reliability Optimization in Large Language Model Fine-Tuning 16
• 9. Normal AI Chatbots 19
• 10. Prompt Engineering in Artificial Intelligence 22
• 11. Fine-Tuning in Artificial Intelligence 25
• 12. Retrieval-Augmented Generation (RAG) 27
• 13. MCP Chatbots 31
• 14. Agentic AI 35
• 15. AI in Low-Resource Data Countries 38
• 16. AI for Education in Developing Countries 42
• 17. AI for Healthcare in Developing Countries 44
• 18. AI Policy and Governance in Developing Countries 47
• 19. AI for Sustainable Development Goals (SDGs) 49
• 20. Research Gap 50
• 3. Methodology 51
• 1. Research Design 52
• 2. System Under Test 53
• 3. Participants and Setting 55
• 4. Instruments 56
• 5. Procedure 56
• 6. Outcome Measures and Hypotheses 57
• 7. Outcome Measures and Hypotheses 58
• 8. Data Preparation 58
• 9. Statistical Analysis Plan 60
• 10. Visualizations and Tables 60
• 11. Ethical Considerations 61
• 12. Tech Stack and Versions 62
• 13. Limitations 62
• 4. Results 63
• 1. Automatic Evaluation (BLEU Score) 63
• 2. Student Survey Results 64
• 5. Discussion 69
• Chatbot Performance (BLEU Evaluation) 70
• Student Survey Results 70
• Student Feedback on Updates 70
• 6. Future Work 70
• 7. Conclusion 71
• 8. References 72
• 9. Appendix 1 76
• 1. Introduction 76
• 2. Global Trends in AI Policy 78
• 3. Current Situation in Laos 80
• 4. Policy Vision and Goals 81
• 5. Strategic Pillars 83
• 6. AI for Education 83
• 7. AI for Healthcare 86
• 8. AI Research and Development 91
• 9. AI Infrastructure 91
• 10. AI Governance and Ethics 92
• 11. Implementation Roadmap (2025–2040) 93
• 12. Monitoring and Evaluation 95
• 13. Conclusion and Policy Recommendations 96