MSc Researcher · Computer Science · UFSC
Daniele
Orzechowski
I study how artificial intelligence can bring specialist knowledge closer to the point of care.
Specialist knowledge,
closer to primary care.
Many Brazilian cities have little or no access to medical specialists. This gap can delay diagnosis and increase patient risk.
My MSc research examines retrieval-augmented generation for clinical decision support in Brazil’s public health system, the SUS. The work asks a precise question: when does better retrieval produce a better answer?
The first case study compares ten retrieval approaches across roughly 18,000 real dermatology case reports. It evaluates retrieval and generation separately, without claiming clinical deployment or patient outcomes.
Core finding / 01
Better retrieval did not improve every generator. Only one of two evaluated language models benefited from retrieved context.
RAG value depends on the complete system, not retrieval alone.Image → representation → clinical language
Learning from
what clinicians see.
A multimodal study tests whether a vision-language model can classify skin lesions and produce a structured clinical-style report.
25,000Dermoscopic images
8Diagnostic categories
71%Classification accuracy
The study fine-tuned LLaMA 3.2 11B Vision and compared eight strategies for a highly imbalanced dataset. The selected setup did not produce broken or invalid output.
Current work
Three connected research systems
Clinical decision support
RAG research that grounds answers in institutional protocols and prior specialist responses.
UFSC · Image Processing and Computer Graphics LabISIS cytology platform
Server infrastructure for neural networks that assess Feulgen- and Papanicolaou-stained slides for oral cancer screening.
UFSC · Telemedicine LabTeleconsultation corpus
A dataset of exchanges between primary care physicians and specialists, with careful patient data de-identification.
SUS · Research in progressWork made
to be examined.
The work spans information retrieval, medical imaging, statistical methods, and human-centered health systems.
Publications03 records
A Comparative Analysis of RAG Architectures for Diagnostic Support Using Telemedicine Data
Orzechowski, Zaniboni, von Wangenheim, Roschildt Pinto & Jeronimo de Macedo
AcceptedSpringer LNAIAwaiting publication
A Proposed Clinical Safety Assessment Framework for Retrieval-Augmented Generation Systems
Methodological synthesis from a systematic review of 65 medical applications
In reviewJournal of Biomedical InformaticsSecond round
Beyond the General Linear Model: Linear Mixed-Effects Modeling of fMRI Data
Orzechowski & Martins da Costa · Human Connectome Project
In reviewNeuroinformaticsSecond round
Selected projectsMethods
Multimodal lesion classification
LLaMA 3.2 11B Vision · classification and report generation
fMRI mixed-effects tool
Validated against the reference R implementation to six decimal places
SIAM medication administration
ESP32 dispenser · Angular · Node.js · PostgreSQL · Docker
Presentations2026
Benchmarking RAG for Dermatological Clinical Decision Support
Beihang University, China
A Comparative Analysis of RAG Architectures for Diagnostic Support
Brazilian Conference on Intelligent Systems
A path across
disciplines and borders.
Software engineering provided the foundation. Artificial intelligence, intelligent systems, and health research now define the direction.
-
2025—27
Florianópolis
MSc in Computer Science
Federal University of Santa Catarina · UFSC
Artificial intelligence in healthcare and machine learning for medical data
Brazil -
2026
Hangzhou
CAPES research mobility
Beihang University
Brazil–China Youth Science & Technology Innovation Leadership Program
China -
2024—26
GPA
Specialization in Artificial Intelligence
Federal University of Technology – Paraná · UTFPR
Machine learning, neural networks, and predictive analytics
9.48/10 -
2024—25
GPA
Specialization in Intelligent Systems and Agents
Federal University of Goiás · UFG
Intelligent systems for complex healthcare solutions
9.44/10 -
2022—24
GPA
Bachelor’s in Software Engineering
Catholic University of Santa Catarina
Machine learning, artificial intelligence, and data science
9.02/10
Research collaboration · Academic exchange · Healthcare AI