Implementation of Language Models for Embedding Generation Applied to Diagnostic Medicine in Kidney Transplantation

• Grant number: 25/12759-9
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: September 01, 2025
• End date: August 31, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: José Fernando Rodrigues Júnior
• Grantee: Roberto Spíndola Abrenhosa Filho
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated research grant: 24/04761-0 - Artificial Intelligence for Improved Infectious Diseases Outcomes in Kidney Transplant Recipients (AIIDKIT), AP.R

Abstract

This undergraduate research project proposes the local deployment of large language models (LLMs) for embedding generation applied to diagnostic medicine in the context of kidney transplantation. The initiative is part of the FAPESP AIIDKIT project (2024/04761-0), which aims to predict infectious risks in transplant recipients. The methodology involves the use of open-source LLMs, textualization of structured and unstructured data, and extraction of high-dimensional semantic vectors. These embeddings will be evaluated in supervised tasks (such as infection prediction) and unsupervised tasks (such as clustering of clinical profiles), with local vector storage using FAISS and PGVector. The project aims to advance the state of the art in medical data vector representation, contributing efficient and interpretable tools to support clinical decision-making, while respecting the computational and ethical requirements of the hospital environment.