M.Sc. student at the University of São Paulo (USP), working with C4AI on Brazilian Portuguese language models. My work spans data engineering, post-training, and model evaluation in projects such as LLMBR, alongside industrial AI and MLOps research in the TRACTIAN–USP collaboration.
|
I build corpora for LLM training through document extraction, normalization, quality filtering, deduplication, tokenization, and synthetic data generation. My work also covers instruction–response datasets for supervised fine-tuning (SFT), plus document preparation, chunking, and indexing for retrieval-augmented generation (RAG).
|
Applied machine learning for condition monitoring and fault detection, combining signal analysis, feature engineering, and reproducible ML systems. Current research explores these methods in servomotor applications, connecting research and software engineering.
|
| Area | Tools |
|---|---|
| ML & data | Python, PyTorch, Transformers, Datasets, NumPy, pandas, scikit-learn |
| LLM engineering | PEFT, TRL, Accelerate, vLLM |
| Systems | FastAPI, Docker, Linux, Git, TypeScript, Node.js, C++ |
 |
Token counting, distribution analysis, and dataset inspection for LLM pipelines. |
 |
Portuguese corpora, instruction datasets, and synthetic-data experiments. |
 |
Implementations and experiments with embeddings, Transformers, BERT, and RLHF. |
 |
Undergraduate research at UFU on scheduling algorithms for AI workloads. |
My background also includes backend development, graph algorithms, and competitive programming.
