Professional experience
I'm a Biomedical engineer and a Master degree student of Applied Computing at UNISINOS since the begining of 2025. I have 2 years of experience at the Startup Biosens Tech, working with manufacturing processses of microfluidic channels of rapid tests for several projects, also having a part in the developing of an automatic system for visual inspection of the biossensors. Academically I have projects in the biomedical image and signal processing field as well as many small projects that can be used as basic generic models for diverse problems for fast prototiping and testing.
As my final paper at college I've developed an algorithm based on neural networks to classify different heart pathologies from an 12-lead ECG raw signal (timeseries classification) trained and tested with the "PTB-XL" database from Physionet using different tools and libraries, such as Pandas, NumPy, MatPlotLib, Seaborn, Keras, TensorFlow, Scykit-Learn, ScyPy, WFDB and OpenCV. Different architectures were tested in different configurations, such as Resnet-50 pre-trained with imagenet images, A CNN-1D made from scratch and an ensemble model of CNN-1D, achieving as a final result 84% of accuracy along the 5 classes.
I am currently a researcher at Biosens Tech, a startup that is developing rapid tests based on biossensors for diferent conditions, such as helping blood coagulation disorders diagnose and monitoring. I am currently responsible for the majoritie of the 2D and 3D models created by the company to improve the manufacturing processes and also responsible for the contact with external suppliers.
Recently the field of data science, I've develop a neural network based algorithm for biossensor non-conformity visual inspection. That uses the Inception V3 architecture to classify different test strips in 5 classes, one for each error and one for test strips that are in good condition. The model trained with more than 5000 images had a accuracy of 96% in the validation dataset and in a test with 16 images it droped to 88%, resulting 14 images correct and 2 images classified incorrectly. Although the model had a lower performance in a production simulation, every test strip with good condition was classified correctly and none of the improper test strips was classified in a good condition, meaning that the production didn't suffer from a significant impact. Currently, the model deployment is in study to research the usability in production.
Personal Summary
I'm a Biomedical engineer and an Applied Computing Master's student. I have 2 years of experience in research at the Startup Biosens Tech, working with manufacturing processses of microfluidic channels of rapid tests for several conditions and a data science project for inspection of production materials. Currently, I'm working at Biopark Educação as an R&D analyst, responsible for R&D KPIs, report automation (involving ETL, data analysis, python developing) and maintenence for current software used in the sector. Academically I have projects in the biomedical image and signal processing and many small projects that can be used as basic generic models for diverse problems for fast prototiping and testing.