EN — I build bridges between public health and machine learning, from one conviction: technology is most powerful when it serves collective needs. I apply LLMs, knowledge graphs and predictive models to real health data, clinical text in Portuguese, patient flow across Brazil's public health system (SUS), and longitudinal cohorts.

PT — Construo pontes entre saúde pública e machine learning, partindo de uma convicção: a tecnologia é mais poderosa quando serve a necessidades coletivas. Aplico LLMs, grafos de conhecimento e modelos preditivos a dados reais de saúde, texto clínico em português, fluxo assistencial no SUS e coortes longitudinais.

• LLMs & GraphRAG — reconstruct patient journeys from health data
• NLP in Portuguese — clinical information extraction
• Efficient fine-tuning (LoRA/QLoRA) and rigorous model evaluation
• MLOps — reproducible pipelines, from raw data to served model

regionalizacao-sus-graphrag — GraphRAG over patient journeys in Brazil's public health system

Knowledge graph in Neo4j plus natural-language-to-Cypher via LLM, using breast cancer as a tracer condition. QLoRA fine-tuning and multi-model evaluation by execution.

Python · Neo4j · GraphRAG · LLM · QLoRA · NLP-PT

who-zscore-python — WHO growth reference z-scores

BMI-for-age and height-for-age z-scores (5–19 years) via the LMS method, validated against the official WHO anthroplus package. Clean, vectorized code.

Python · pandas · numpy