Giselle: A RAG-Based Generative AI Platform for Mental Health Care
Fabio Jose Gomes de Sousa1, Ronaldo Fernandes Ramos2, César Olavo de Moura Filho2, Ivana C. H. Cunha Barreto3, Luiz Odorico Monteiro de Andrade3 and Antonio Mauro Barbosa de Oliveira2
1 Federal Institute of Ceará, Maracanaú, Ceará, Brasil (Email: )
2 Federal Institute of Ceará, Fortaleza, Ceará, Brasil (Emails: , , )
3 Fundação Oswaldo Cruz, Eusébio, Ceará, Brasil (Emails: , )
DOI: https://doi.org/10.48545/advance2026-shortpapers-3_4
Abstract
The launch of GPT-3 by OpenAI in 2020 revolutionized the field of Natural Language Processing (NLP) by introducing highly advanced generative AI models capable of performing complex tasks. This breakthrough accelerated the global adoption of artificial intelligence in various fields, including automation, customer support, and content creation. However, in the healthcare sector, the use of generative models requires caution, as errors or “hallucinations” can have serious consequences. A promising approach to improving the safety and accuracy of these tools is Retrieval-Augmented Generation (RAG), which enables the model to access external information in real time, ensuring up-to-date content, solution customization, and greater transparency—without the need for retraining—while significantly reducing the risk of hallucinations. In this context, this study explores the application of RAG in the Giselle Saúde project, an initiative funded by Embrapii/MCTI, which leverages generative AI to detect signs of depression in the elderly.