Alertawy | AI-Powered Driver Monitoring System

Alertawy is a comprehensive driver monitoring system designed to address road safety concerns and reduce accidents caused by driver distractions. Using advanced deep learning techniques, it can accurately detect signs of drowsiness and distraction, providing alerts to prevent potential accidents.

AI-Generated Images Detection

End-to-end full-stack AI application to distinguish AI-generated and human-created images using advanced deep learning, custom bias-reducing loss, and extensive data augmentation. Deployed on AWS via containerization (Docker) and CI/CD (GitHub Actions). Ranked top 20/550+ on its Kaggle Competition.

Bondora Financial Risk Predictions

Developing a credit risk model for peer-to-peer lending using Bondora system. Includes data cleaning, EDA, feature engineering, building classification/regression models, and deploying the models to a web app for evaluating borrower creditworthiness.

Data Analysis for the High School Final Exams in Egypt

Cleaning scraped datasets for Thanawya Amma exams, exploring and visualizing data by specialization using Plotly, matplotlib, and Seaborn. Insights gained on student grades, informing educational policies for better performance, and Helping new students gain insights about Thanawya Amma exam grades.

Real-Time Violence Detection

Proposed a deep learning model that uses a MobileNet Pretrained Model as a spatial feature extractor and Bidirectional LSTM as a temporal relation learning method. The model achieved high accuracy (97%) and fast response time (16 frames/sec), making it effective for real-time security applications.

Diamond Prices Prediction Competition

Predicting diamond prices, We explored data through comprehensive analysis and visualization. We then extensively cleaned and preprocessed it, creating an efficient preprocessing pipeline. We experimented with various machine-learning models. Remarkably, we achieved a top 10 ranking among 70 teams.

COVID Chest X-ray Detection Competition

Classify Chest X-ray images. Started with a Vanilla CNN model, then incorporated pre-trained models like DenseNet, InceptionNet, and VGG through transfer learning. Extensive experimentation led to improved performance and secured the third position among 30 teams.

Movie Reviews Sentiment Prediction Competition

Predict movie review sentiment. Explored, cleaned, and preprocessed data using text-cleaning techniques. Developed ML boosting model and DL models (CNN, Bidirectional LSTM). Created prediction pipelines with ML/DL models and deployed them with FastAPI. Achieving top 10 positions among 20 teams.

Music Genre Classification Competition

In this competition, our goal was to predict the genres of a music tracks dataset, we performed thorough data exploration, and comprehensive feature engineering and extraction, Then we experimented with various machine-learning ensemble and boosting models, Securing a top 5 position out of 50 teams.

Customer Segmentation Credit Card Analysis

Utilized unsupervised learning techniques PCA, and K-means to segment customers based on credit card purchase behavior. The analysis involved dimensionality reduction, cluster analysis, optimal cluster determination, and visualizations. Findings provided insights for targeted marketing strategies.