Development of a Privacy-Focused Khmer Chatbot Based on Retrieval-Augmented Generation

RAG 기반 개인정보 보호형 크메르어 챗봇 개발

Khmer language processing faces significant challenges due to limited resources, making it difficult to build effective AI systems. This study proposes a privacy-focused Khmer chatbot using a Retrieval-Augmented Generation (RAG) framework, designed to operate entirely within secure infrastructures without exposing sensitive data to external servers. A local FAISS-based database stores internal documents, enabling accurate and private retrieval. We evaluated five pre-trained large language models (LLMs) and two embedding models to select the optimal configuration for Khmer question answering. The final system improves both response accuracy and information privacy, providing a scalable solution for private organizations and contributing to Cambodia's digital transformation.