Hi, I'm Mohamed Abdelnaby

×

RSA Algorithm

During my high school studies, I initiated a comprehensive exploration into the RSA cryptography, a robust public-key cryptosystem. My project delved into the core processes, such as key generation, encryption, decryption, and the pivotal prime factorization problem within RSA. In addition to the theoretical aspect, I applied the concepts by developing a Python program, demonstrating the RSA algorithm’s application. This practical implementation encapsulates the generation of keys, encryption, and decryption of messages, serving as a practical demonstration of RSA's operations. The study concluded by considering the future implications of advancements in technology on cryptography, including the potential impact of NP=P and the development of quantum-resistant symmetric cryptography. This inaugural research experience allowed me to develop a deep understanding of the RSA system and set a foundation for future investigations in the field.

×

IDS using Machine Learning

In my freshman year of college, I led a project on Network Intrusion Detection using Machine Learning (ML). This included collecting network logs, preprocessing data, and selecting relevant features. I applied machine learning models, such as Naive Bayes, Decision Trees, and Random Forests, to categorize network traffic as normal or intrusive. The models demonstrated high detection rates, improving our ability to identify potential network threats. The project concluded with an analysis of results and outlined potential directions for future research. This project contributed significantly to my understanding of ML applications in cybersecurity.

×

Adversarial ML Algorithms and Detection

This is ongoing project and is set to be published, please reach out to me for more details. Here's an overview of the project:

The aim of this project is to investigate the existing Adversarial Machine Learning algorithms including but not limited to FGSM (Fast Grident Sign Method), DeepFool, C&W, PGD (Projected Gradient Descent), and BIM (Basic Iterative Method)

These Adversarial Attacks are evaluated on a wide range of Machine Learning Models including Liner Regression, Logistic Regression, Decision Trees, Support Vector Machine, K-Nearest Neighbors, Random Forest, and Naive Bayes

Then, we aim to develop ways to run these Adversarial Machine Learning algorithms effectively on these ML models, as we are going to investigate techniques of detections

Moreover, in this project, we also aim to rethink the definition of detectability

×

Urban Scaling

This is an ongoing project and is set to be published, please reach out to me for more details. Here's an overview of the project:

Over 500,000 Latin inscriptions were scrapped and appended to a database. These Latin inscriptions were extensively cleaned and analyzed

We used BeautifulSoup and Requests libraries to gather and extract the inscriptions. Pandas was used to clean and perform data manipulation techniques (merging, grouping, reshaping, pivoting, etc.) on the Latin inscription text

This database was then merged with another database that contained key information about provinces and cities

CLTK library was used for Latin Language Lemmatization and Synonymization

We introduced a technique to gather all the latin words extensions and to run a frequency matching function for the entire database

Applied clustering techniques and used algorithms such as KBinsDiscretizer, Decision Tree, K-means Clustering, DBSCAN, Optics Clustering, Gaussian Mixture Model, Agglomerative Clustering, and Hierarchical Clustering

Applied Statistical analysis and visualization using libraries like Matplotlib and Seaborn.

Scikit-learn was used to further the data analysis

Linear Regression model was used for Urban Scaling

×

Advanced AI Chatbot

Inspired by the capabilities of sophisticated AI models such as ChatGPT, GPT-3, and GPT-4, a robust, domain-specific chatbot "itsupport.ai" was engineered to handle IT support queries at the University of Oklahoma. The chatbot was designed to bridge the knowledge gap observed in existing AI models when dealing with unique, repetitive queries specific to the university's IT support context. The chatbot operates through an interactive interface where user questions are channeled to the backend to generate an appropriate response. The system utilizes a customized index derived from an extensive library of over 380 IT support articles from the university. Leveraging this resource, the GPT-3 model is primed with contextual information from this index, enabling it to generate accurate responses. The model is hosted on an AWS server and is accessed through a FastAPI Python framework for streamlined operation. Despite facing challenges related to server maintenance on an AWS EC2 instance, designing a frontend with conditional rendering, collecting and preparing data through web scraping, fine-tuning the GPT-3 model, and managing user-interaction on the frontend, successful solutions were devised and implemented. This accomplishment is a testament to the robust backend infrastructure, the user-friendly and responsive frontend design, the high-quality data collection and preparation process, and the effective fine-tuning of the GPT-3 model. The development process cultivated skills in deploying API servers on AWS EC2, customizing GPT-3 with specific data using llama-index, understanding Large Language Models, interfacing with OpenAI's APIs, and state management in React with API requests. Looking ahead, there are ambitious plans to refine and enhance the chatbot. This includes expanding the dataset, integrating user feedback, adding multi-language support, integrating with other support channels, personalizing user experiences, offering voice-enabled support, and incorporating analytical and reporting mechanisms to monitor performance and facilitate continuous improvement. The chatbot is primed to become an invaluable asset in the university's IT support framework.

×

Hacklahoma's website & Registeration System

The Hacklahoma Website & Registration System uses PostgreSQL and a private virtual machine to efficiently handle the Hackathon's needs. PostgreSQL, a reliable database system, manages participant data securely during the registration process. The website and registration system are hosted on a private virtual machine, which ensures control, performance, and security. The registration system is designed for user-friendliness, with participants' details easily input and managed. The website, the main hub for Hacklahoma's 2024 Hackathon Season information, maintains engagement within the University of Oklahoma community.

×

AutoGuardian: Car Maintenance React Native App with supabase

This app was developed during the HackUTA 2023 Hackaton. This was my first experience with mobile development. This app has multiple features that are outlined below:

User Authentication: Using Supabase with a Postgres database, we established a system for user sign-up and account creation.

Maintenance Scheduler: This feature enables users to map out and track regular car upkeep routines.

Knowledge Hub: A section dedicated to car maintenance tips and tricks.

Maintenance Log: Users can record and monitor their car's maintenance history over time.

Reminder System: Our application pushes timely notifications, alerting users about impending car maintenance tasks based on their prior inputs.

What sets our app apart is the integration of an advanced chatbot. We tapped into the capabilities of OpenAI's GPT model, fine-tuning it via the API to cater to queries specifically in the car maintenance sphere. This ensures our users receive informed and relevant responses.

×

Message Status