@mastersthesis{design-gsi-mastersthesis-20265,
author = "Javier S{\'a}nchez Mayoral",
abstract = "The aging rate of the global population is growing at an unprecedented rate, increasing the prevalence of frailty syndrome, a clinical condition whose early identification is challenging due to overlapped symptoms with normal aging. This underscores the need to provide the population with understandable medical information and to offer clinical personnel with rapid access to specialized knowledge. Although Large Language Models (LLM) offer promising possibilities to process natural language queries, their tendency to hallucinate and static knowledge limit their deployment in clinical settings. To address these issues, this Bachelor’s Thesis presents the design and development of a conversational system specialized in frailty syndrome based on the Retrieval-Augmented Generation (RAG) paradigm.
The system relies on a specialized knowledge base indexed using clinical guidelines, medical literature, and educational material aligned with the project’s scope. The core innovation of this project is the implementation of an adaptive mechanism to the user profile (patient or healthcare professional), which influences both the retrieval and generation pipelines. Furthermore, the system supports conversation history by implementing a query reformulation stage and includes a transparency panel that exposes the source of the retrieved contexts.
System performance was evaluated through a sequential ablation study using a testset of 85 interactions, combining quantitative metrics and domain-specific qualitative clinical rubrics. Experimental results demonstrated that the optimal configuration of the system is made up of the qwen3-embedding model, a hybrid metadata vector retrieval strategy with k = 12, and the llama3.3:70b generation model, which maximizes the balance between precision and recall and minimizes hallucinations by achieving high context faithfulness scores. Ultimately, this prototype establishes a solid foundation for the delivery of personalized and traceable digital health solutions within the geriatric care domain.
",
institution = "Universidad Polit{\'e}cnica de Madrid",
title = "{D}esign and {D}evelopment of a {RAG}-based {C}onversational {S}ystem for {F}railty {S}yndrome {S}upport",
year = "2026",
}