Myanmar Agri-Med AI: Developing an Instruction-Tuned Burmese LLM for Rural Knowledge Dissemination

Read full text Summarise with VIRAT
Authors: Phone Min Khant
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Abstract— The lack of reliable, domain-specific AI tools in the Burmese language creates a digital divide for rural communities in Myanmar. To address this problem, this paper presents Myanmar Agri-Med AI, an instruction-tuned conversational AI designed specifically for the agriculture and medicine sectors. We fine-tuned a lightweight multilingual model (paraphrasemultilingual-MiniLM-L12-v2) using Burmese datasets, including "Burmese-Microbiology-1K" and a filtered agriculture dataset. To ensure the AI provides safe and factual advice, the system uses a Retrieval-Augmented Generation (RAG) architecture with FAISS (Facebook AI Similarity Search). This method retrieves answers only from verified knowledge bases, preventing the AI from generating false information. The final solution is hosted on a simple Streamlit web interface that supports Unicode Burmese input, domain switching, chat history management, and automatic medical disclaimers. System testing shows that the model achieves a 90% semantic match accuracy and a fast response time of 0.85 seconds. Furthermore, a safety distance threshold effectively blocks out-of-domain queries, and user evaluations resulted in a high overall score of 4.67 out of 5 for linguistic clarity, safety, and usability. This project demonstrates that low-resource Natural Language Processing (NLP) techniques can successfully create socially meaningful tools to support rural health and farming.

Fields of Study & Subjects

Topic Modeling Natural Language Processing Techniques Artificial Intelligence in Healthcare and Education

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
30 March 2026
Language
EN