Wanderson souza

Hi, I'm Wanderson

​​

I’m a Machine Learning Engineer and Data Scientist with a Ph.D. in Mechanical Engineering, an M.Sc. and a B.Sc. in Computer Science. My AI journey began in academia, where I built a computer vision system to estimate metro passenger flow. This was my first real-world application of deep learning and predictive modeling.

At NeoPTO, I trained hierarchical NLP models for patent classification, processing over 12 million records and supporting systems for semantic search and trend analysis.

In 2023, I joined iTriad (now Venturus), where I led computer vision projects for embedded systems, including head pose estimation, the Merge Model (a unified detector for body, face, pets, and dolls), and facial recognition using infrared imagery. I worked with architectures like ViT, ResNet, MobileFaceNet, and YOLO, and loss functions including ArcFace, Triplet Loss, CosFace, Focal Loss, Center Loss, and Contrastive Loss.

Currently, I lead the AI Security project focused on liveness detection (anti-spoofing) for both RGB and IR modalities, with the goal of achieving iBeta Level 1 and Level 2 certification. I manage the full pipeline from planning to deployment. I also lead the technical alignment with dataset vendors, defining data requirements and evaluating sample quality before acquisition. In addition, I coordinate certification efforts through direct meetings with the iBeta team to clarify protocols and ensure ISO/IEC 30107-3 compliance.

I developed a certification simulator based on the iBeta standard and implemented a complementary CNN-PCA-SVM strategy to enhance decision confidence by increasing feature diversity. 

I’m passionate about building robust AI systems that move seamlessly from research to production, with a focus on performance, generalization, and deployment efficiency.