What does this research mean for the field?

An intelligent service system integrating generative AI, retrieval-augmented generation, and multimodal interaction provides high-quality, fluent, and contextually grounded consultation support in hospital environments. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To design and implement an intelligent service system integrating generative AI for hospital consultation support.

May 31, 2026Open Access

A generative AI–enhanced intelligent service system with contextual retrieval and adaptive interaction for hospital use

Key Points

To design and implement an intelligent service system integrating generative AI for hospital consultation support.
Developed a modular AI service architecture integrating generative AI and contextual retrieval.
Utilized voice activity detection and speech-to-text modules for stable user input capture.
Conducted experimental evaluations in collaboration with a regional hospital under controlled conditions.
Achieved BLEU-4 score of 0.71, indicating high language generation quality.
ROUGE-L score of 0.83 reflects strong content recall in generated responses.
Fluency score of 4.6/5 demonstrates high user engagement in interaction evaluation.

Abstract

This study presents the design and implementation of an intelligent service system that integrates generative AI, contextual retrieval, and adaptive interaction mechanisms to support frontline consultation in hospital environments. Motivated by the challenges faced in HIV and infectious-disease counseling—such as limited manpower, varying patient needs, and the demand for timely and standardized information—this research proposes a modular AI service architecture capable of delivering context-aware and empathetic responses. The system combines multi-stage speech interaction, Retrieval-Augmented Generation (RAG), and domain-adapted prompting strategies to support contextually consistent and retrieval-grounded responses in medical consultation scenarios. A distance-sensing trigger module initiates the interaction, while Voice Activity Detection (VAD) and speech-to-text (STT) modules capture user speech input with stability. The RAG module searches a curated medical knowledge base to provide domain-relevant contextual information, and the large language model (GPT-4o-mini) generates contextually grounded and comprehensible responses. The system’s multimodal output—consisting of synchronized text, speech, and animated visual feedback—supports user engagement and interaction fluency. Experimental evaluation conducted in collaboration with a regional hospital demonstrated positive results in language-generation quality and exploratory usability assessment under controlled or semi-controlled evaluation conditions (BLEU-4 = 0.71; ROUGE-L = 0.83; fluency score = 4.6/5). Scenario-based evaluations further suggest that the prototype can support retrieval-grounded consultation workflows and provide traceable interaction records under the evaluated conditions. Overall, the proposed system demonstrates the feasibility of integrating retrieval-grounded language generation, multimodal interaction, and embedded sensing into a human-centered consultation-support framework for healthcare service environments.

Mark Helpful

Bookmark

Relay

View Full Paper