Abstract

Despite advances in public policies aimed at the LGBTQIAPN+ population, there is still a lack of consistent data on this community – particularly on trans individuals – due to social visibility and a history of prejudice. This article presents the development of Vivi, a virtual assistant designed to support and guide trans people who are victims of violence, fostering connections with support networks and specialized services. The methodology included a literature review, automated data collection through web scraping from public sources, preprocessing with Python, and the construction of a vector database using FAISS with embeddings generated by the paraphrase-MiniLM-L6-v2 model. Response generation was carried out using CEIA-UFG/Gemma-3-Gaia-PT-BR-4b-it-model, integrated into the RGA architecture. The interface was developed with Streamlit, incorporating visual elements inspired by the trans flag. Results demonstrate the technical feasibility and social potential of the solution, although validation with real users is still required to confirm its effectiveness.

Download

Share

COinS