Aplicação de processamento de linguagem natural como ferramenta para suporte à gestão farmacêutica em ambiente hospitalar

Abstract

Hospital pharmacy management involves significant volumes of data and requires rapid access to reliable information to support decision-making processes. This methodological study aimed to explore the application of Natural Language Processing (NLP) in hospital pharmacy management through the development of a chatbot prototype based on a Large Language Model (LLM) from the Qwen family. The research was conducted in two complementary phases: the first, exploratory phase, dedicated to identifying the main information requests related to inventory management, and a second phase of development and functional evaluation of the prototype, implemented in Python. The evaluation of the tool focused on a structured set of questions representative of the common needs of professionals, resulting from the analysis of participants' responses during the first phase. The results showed an overall execution success rate of 85%, indicating that the model was able to generate valid SQL queries in most cases. Performance proved satisfactory for simple questions of low to moderate complexity, which produced consistent and actionable responses. However, a significant proportion of more complex queries did not lead to semantically correct answers, despite successful technical execution, mainly highlighting limitations in the interpretation of implicit business rules. Overall, the prototype demonstrated technical feasibility and potential for practical application, while also emphasizing the need for methodological adjustments and model refinements to improve the accuracy of responses in more complex decision-making scenarios.