• Building multimodal machine learning systems for healthcare
• Researching cross-attention mechanisms in vision-language models
• The impact of VLM analysis of skin lesion images.
• Study of juridic sentences' classification process (Embedding sentences for juridic text classification).

• Chatbot integrated with TELEGRAM bot: A simple chatbot integrated with TELEGRAM bot with support for English, French, Spanish, and Portuguese. Additionally, the history of user conversations is stored in a MongoDB (NoSQL) database; however, users can delete it without losing their registration. The last upgrade can read and then analyze a PDF file according to the user's request for the LLM model. (https://github.com/wyctorfogos/telegram-chatbot-and-object-description)
• NER: Recognition of Entity using a BERT-based model to anonymize textual sentences. NER
• Clothes Detection: A Dockerized microservice that detects different types of clothing via a REST API.
• Chatbot Service: A containerized chatbot deployment for real-time user interaction.
• Safe Vestments Detection: A security supervisor tool to verify the proper use of safety gear. Watch Demo
• JWT Authentication API: Secure service for issuing, registering (PostgreSQL), and invalidating JWT tokens.
• Dijkstra algorithm: An application of Dijkstra. (https://github.com/wyctorfogos/Dijkstra-algo-project-data-structure.git)
• License Plate Reader: Image preprocessor that reads and recognizes license plate digits.
• Concrete Crack Recognition: Model and interface for identifying cracks in concrete structures.
• Crop Faces using MTCNN: Docker microservice to detect, crop, and preprocess human faces.

Explore more: My Repositories

• Benchmarking SRGAN-Upscaled YOLO for Enhanced Object Detection in Aerial Imagery This paper addresses object detection in the domain of aerial images, a challenge due to the large diversity of images concerning the image quality and the varying sizes of objects within them. A training and evaluation strategy based on a Super-Resolution Adversarial Generator Network (SRGAN) model is proposed. The latter generates several new images of higher quality than some of the images available in the dataset to replace them. This strategy increases the learning of the model to better detect objects at a great range of scale sizes, which increases the performance of the YOLO model family. Our extensive benchmarking study on the DOTA v1.5 dataset compares ten YOLO architectures across three evaluation scenarios at input resolutions of 416×416 and 640×640 pixels. Experimental results demonstrate that YOLOv5s-transformer trained on SRGAN-upscaled images (scale factor 2) achieves superior detection performance, with up to a 3.07% absolute increase in mAP0.5 and a 2.29% gain in mAP0.5:0.95, respectively, when compared to conventional training approaches. This study establishes benchmarks for future comparisons with emerging YOLO variants and alternative super-resolution methods.

• MetaBlock-SE: a method to deal with missing metadata in multimodal skin cancer classification Usage of patient metadata (sentence text) on VLM models to classify skin lesions.

• Systems Analysis: Introduction to System Functioning
  Fundamentals of systems analysis, modeling, and documentation.

• Introduction to Production Planning and Control
  Course support material for the logistics technical program.

• Uso de IA na Construção Civil
  Benchmark study on anomaly detection in civil structures using machine learning.

• Gestão de Estoques
  Innovative approaches to inventory optimization.

• Uso de IA na Indústria Brasileira
  Literature review on AI applications in the Brazilian industry.

• Uso de GAN para melhora da detecção de objetos em imagens aéreas(Super-Resolution GAN Improving YOLO's Performance Benchmark) International Conference IEEE - ICECCME 2025 and ICMLA 2022

• Diagnóstico e Reabilitação com Realidade Virtual
  VR system for diagnosis and rehabilitation of lower-limb amputees.

• Gait Movement Recognition using CNN & StarRGB
  Space-Time Information Condensation Technique for Remote Physiotherapy

• Automatic Lighting Control System
  Lighting automation based on presence sensors and Bluetooth.

Languages & Frameworks: Python, JavaScript (Node.js, Express), C, SQL (PostgreSQL, MySQL)
Web & Backend: HTML5, CSS, FastAPI, Flask, JWT, Nginx (Reverse proxy) Data & ML: TensorFlow, PyTorch, scikit-learn, OpenCV, Matplotlib, Pandas, Seaborn DevOps & Tools: Docker, Kafka, RabbitMQ, Grafana, Redis, MongoDB, Git, Postman, AZURE