A Domain-Specific Retrieval-Augmented Generation Chatbot for Medical Queries on Adenomyosis and Endometriosis

Abstract

This study introduces a domain-specific Retrieval-Augmented Generation (RAG) chatbot for answering medical queries on adenomyosis and endometriosis in Bahasa Indonesia. The system employs a rewrite-retrieve-read paradigm, multilingual retrieval, and citation-aware generation using the Mixtral model. Document chunking was optimized for better alignment between user queries and source texts. Evaluation with the RAGAS framework yielded a faithfulness score of 72.92 and context recall of 62.09, with lower performance in answer relevancy (50.23) and context precision (44.78). These findings indicate effective grounding yet reveal the need for retrieval refinement. By adopting automated, reference-free evaluation, this work highlights both the limitations and the potential of domain-adapted RAG systems, contributing a scalable and explainable solution for medical QA in low-resource languages.

Ähnliche Arbeiten