Scholarlink

In this paper, we present a conversational AI system that addresses the ambiguity of patient reported symptoms in medical triage. Designed for voice interactions in Moroccan Darija, the system begins by getting patient speech via a DeepFilterNet2 and Wav2vec2 pipeline. The core of our work is a subsequent Retrieval Augmented Generation (RAG) dialogue engine. After the initial translation of the patient query, this engine retrieves contextually similar past cases from a vector store. This retrieved data is then leveraged by an Atlas Chat LLM 9B, which serves as the central reasoning component, to formulate and ask clinically relevant follow-up questions in the patient’s native language. By expanding the initial query into a comprehensive dialogue, the system provides a high-quality, enriched input to our specialized BERT model for final classification. This RAG-driven, conversational approach significantly enhances the quality of input for classification, resulting in more reliable triage and facilitating the creation of structured Electronic Medical Records (EMRs).

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Advanced Conversational AI System for Medical Specialty Triage Featuring a RAG-Based Dialogue Engine

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Anas Chahid,
  • Ismail Chahid,
  • Mohamed Emharraf,
  • Mohammed Belkasmi

摘要

In this paper, we present a conversational AI system that addresses the ambiguity of patient reported symptoms in medical triage. Designed for voice interactions in Moroccan Darija, the system begins by getting patient speech via a DeepFilterNet2 and Wav2vec2 pipeline. The core of our work is a subsequent Retrieval Augmented Generation (RAG) dialogue engine. After the initial translation of the patient query, this engine retrieves contextually similar past cases from a vector store. This retrieved data is then leveraged by an Atlas Chat LLM 9B, which serves as the central reasoning component, to formulate and ask clinically relevant follow-up questions in the patient’s native language. By expanding the initial query into a comprehensive dialogue, the system provides a high-quality, enriched input to our specialized BERT model for final classification. This RAG-driven, conversational approach significantly enhances the quality of input for classification, resulting in more reliable triage and facilitating the creation of structured Electronic Medical Records (EMRs).