I'm Marlon Sousa

A computer vision system to detect and recognize faces in real-time, using Convolutional Neural Networks (CNNs) and techniques such as OpenCV for image capture and processing. The project can be expanded to include facial authentication systems or security monitoring.

Development of an algorithm that tracks objects in videos using techniques such as optical flow algorithms or convolutional neural networks for object segmentation. The project can be applied to various areas such as security surveillance, motion analysis, or activity monitoring.

Numerical Detector is an interactive application built with Streamlit and TensorFlow/Keras that allows users to draw handwritten numbers and receive predictions from a trained model using the MNIST dataset. The intuitive interface provides a hands-on experience in testing the model's ability to recognize handwritten digits, making it a great tool for exploring Computer Vision.

GenDigit is an artificial intelligence model based on generative adversarial networks (GANs) trained to generate realistic images of handwritten digits. Using a generator conditioned on specific numbers, the project enables the creation of customized samples for pattern recognition applications, data augmentation for model training, and experimentation with generative neural networks.

This project focuses on leveraging Large Language Models (LLMs) for the automated generation of security reports from web server logs. The system is designed to analyze extensive web server logs, extract relevant security-related information, and produce human-readable reports that provide insights into potential security incidents, patterns of malicious activities, and overall server health.

This project aims to predict phishing cyber attacks using BERT and NLP techniques. By analyzing email content and URLs, the model identifies phishing patterns with high accuracy, enhancing cybersecurity defenses against evolving threats. BERT's contextual understanding significantly improves phishing detection over traditional methods.

Paper.Not is an innovative project that leverages large language models (LLMs) to revolutionize the way users interact with scientific papers. The platform allows users to upload a scientific paper and receive an instant, concise summary generated by an advanced LLM. This summary distills the paper’s key points, helping readers quickly grasp its core ideas without having to read through the entire document.

This project develops a personalized music recommendation system using Large Language Models (LLM). It generates curated playlists based on users' preferences, emotions, and listening history. By integrating LLM's natural language processing with Spotify's music catalog, the platform delivers dynamic, personalized music recommendations, aiming to enhance the listening experience by adapting to individual tastes and moods.

The Sentiment Analyzer is an advanced Natural Language Processing (NLP) project designed to classify the sentiment of text data using BERT (Bidirectional Encoder Representations from Transformers). This tool leverages state-of-the-art machine learning techniques to analyze and interpret the emotions or opinions conveyed in text, determining whether they are positive, negative, or neutral.

This project applies reinforcement learning (RL) to train an AI agent to autonomously play Super Mario Bros. The agent learns through trial and error, optimizing its actions to maximize rewards. Using RL methods like Q-learning or deep Q-networks (DQN), the agent balances exploration and exploitation, improving gameplay by learning strategies to efficiently complete levels and overcome obstacles.

An agent trained with Deep Q-Network (DQN) to solve the classic CartPole problem, where the goal is to balance an inverted pendulum on a cart moving left or right. This project demonstrates fundamental reinforcement learning concepts, including exploration vs. exploitation and learning optimal policies.

Development of an agent trained with Q-Learning or SARSA to find the exit of procedurally generated mazes. The agent learns to navigate the environment, avoiding dead ends and optimizing its trajectory for the most efficient solution. This project highlights the application of RL in navigation and path planning environments.

Development of a machine learning-based system to identify and predict fraudulent transactions on e-commerce platforms. The model analyzes purchasing behavior patterns, such as amounts, location, and user history, to detect anomalies in real-time. Algorithms like Random Forest and XGBoost can be used for classification and prediction, helping to prevent financial losses and increase online transaction security.

Creation of an artificial intelligence application that processes web server logs to generate automated security reports. The system uses log analysis techniques and machine learning to detect patterns of anomalous traffic, intrusion attempts, and security vulnerabilities. The project can be expanded to include recommendations on how to mitigate risks and improve server security, using approaches such as NLP and clustering to classify critical events.

Development of a predictive model to detect phishing attacks in real-time, focusing on fraudulent emails, websites, and messages. Using supervised learning, the system analyzes data such as textual content, URLs, and metadata to classify and identify phishing attempts. Techniques like sentiment analysis and feature extraction can be employed to improve detection accuracy, providing an additional layer of protection against online scams.

The BB84 protocol is a Quantum Key Distribution (QKD) scheme developed by Charles Bennett and Gilles Brassard in 1984. It enables two parties, traditionally called Alice and Bob, to securely share a secret key, ensuring resistance against eavesdropping attempts by an adversary (Eve).

QuantumTeleportViz simulates and animates quantum teleportation, a phenomenon where a quantum state is transferred from one qubit to another without the physical transmission of information, using quantum entanglement.

LightCaster 2D is an interactive real-time light propagation simulator that visually demonstrates the fundamental principles of geometric optics in a two-dimensional environment. This educational application allows users to explore how light propagates in a straight line and interacts with different obstacles, creating illumination and shadow patterns.

WaveDuality is a 2D interactive simulator in C++ and SFML that demonstrates the famous double-slit experiment, one of the pillars of quantum physics that reveals the dual wave-particle nature of matter. This software allows real-time visualization of interference phenomena, providing a valuable educational tool for students and physics enthusiasts.