Welcome to my Portfolio!

Open Source Tool: RetinaFace

The RetinaFace tool provides precise detection of the facial region along with five key landmark points: the coordinates of the right and left eyes, the tip of the nose, and the corners of the mouth. This information is essential for tasks such as facial alignment, emotion recognition, and biometric analysis. By extracting these reference points accurately, RetinaFace enables a wide range of algorithmic applications in computer vision, particularly in face-based detection, classification, and real-time facial tracking.

Open Source Tool: DLib

The DLib library provides a highly accurate facial landmark detection system, returning 68 key points across various facial regions, including the eyebrows, eyes, nose, mouth, and jawline. These points are essential for understanding facial geometry and enable tasks such as facial expression analysis, head pose estimation, and feature tracking. A demonstration GIF is included where the tool is applied to my own face, showcasing the real-time adaptation of the landmarks as the face moves. This dynamic tracking highlights the robustness and precision of DLib, making it a powerful open-source tool for numerous computer vision applications involving facial recognition, emotion detection, and gesture analysis.

Microservices for Vibration Monitoring Application - LMEst

Integrating monitoring systems in industrial environments is challenging due to diverse protocols, data formats, and existing infrastructure. This requires customized solutions tailored to each company's needs.

Microservices offer a modular approach, enabling independent, scalable components that adapt to various industrial requirements. This work presents a microservices system using the AMQP protocol to create an easy-to-integrate communication bus.

The system reads data from multiple sensors and provides uniform, accessible data integration for industrial systems. It was implemented and tested in an industrial-like environment with two sensors monitoring a rotating machine's performance continuously.

Results show that the microservices architecture combined with the AMQP bus delivers efficient, flexible integration, addressing the specific needs for monitoring and data analysis in industrial settings.

For more details, visit Zenodo Repository .

Personal Object Labeling and Detection

It is possible that, from Detectron2, reusing the CNN training in the coco-dataset, perform fine-tuning and obtain personalized object labeling and detection, which is fundamental for specific applications that require precision in identifying unique and distinct categories that are not present in generic datasets, thus allowing more adapted and efficient solutions for real problems in different sectors.

Data Science Specialization Project

Failure Prediction System for Rotating Machines Based on Engineering Data and Vibration Signals

This ongoing project is part of my specialization in Data Science at XP Educação (IGTI), and aims to develop a predictive maintenance system for rotating machinery using both engineering data and vibration signal features.

The project is based on the publicly available Kaggle Wind Turbine Signal Vibration Dataset , which contains vibration measurements from wind turbines under various operational conditions.

The pipeline involves data cleaning, feature extraction in the time and frequency domains, and machine learning model development using algorithms such as Random Forest, SVM, and Gradient Boosting. The models are evaluated based on precision, recall, and F1-score for fault prediction.

The goal is to build a scalable solution that can identify early signs of mechanical failure, helping reduce downtime and improve maintenance planning in industrial environments.