Hi there

I'm Mohamed, a software developer and a junior DS/CS college student. I have passion for all things technology and design related, from software engineering & ai/ml to robotics & 3D graphics

Beyond my love for tech and design, I'm interested in religion, history, philosophy, film, and martial arts.

Below are some of the projects I have built/working on.

Projects

Noded

2024-Present

• React
• TypeScript
• Sql
• AI
• Web Dev

Noded is a dynamic mind-mapping tool designed for intuitive thought organization and visualization. It allows users to create multiple interconnected maps, each featuring customizable categories and color-coding for better content management and organization.

The inspiration for this app came from my own experience of shifting through countless screenshots and notes related to organizations or individuals I wanted to reach out to or communicate with. Over time, I often found myself forgetting why I had saved certain information or what its purpose was. This led me to develop a structured way to keep track of such details more effectively. Currently, I am working to expand this idea into a more comprehensive and well-rounded app.

Peeko

2024

• TypeScript
• Python
• Web Dev

Peeko is designed to transform the way academic research could be navigated. By breaking down complex research papers into manageable insights and presenting them in engaging, interactive formats, this project aims to make research exploration more accessible.

At first I was planning to just make an instagram stories like app for resaerch papers, but I'm also trying to expand more on this.

Trespassing Detection

2024

• Python
• AWS
• OpenCV

This project is a Kinesis Video Streams processor that identifies potential intruders in real-time. It captures video fragments from an AWS Kinesis video stream, extracts frames, and uses Amazon Rekognition to detect people. Detected instances are logged to DynamoDB for tracking and the corresponding frames are uploaded to S3 for storage.

Nidaa

2024

• Web Dev
• TypeScript
• Signal Processing
• Pattern Recognition
• Secure Communication

A SOS web app designed for war zones and disaster areas, providing autonomous emergency detection and communication when traditional networks fail. It leverages device sensors and smart response protocols to detect dangers like explosions or building collapses. Inspired by the challenges faced by our people in Gaza and South Lebanon, it operates entirely offline with zero data collection for privacy and avoiding worst case scenarios.

---notifications aren't showing in this recording, will update soon

RPlace Vision Transformer

2024

• Python
• SQL
• PyTorch
• ViT
• Data preprocessing
• Experimentation with model architectures

This work is inspired by an existing project that explored similar ideas. However, I am reconstructing it from the ground up, incorporating my own approach and adjustments to methodology. This project examines the dynamic evolution of the 2022 Reddit Place canvas by analyzing a massive dataset of over 160 million pixel edits, which includes details like color, coordinates, timestamps, and anonymized user information. The dataset is unordered and requires extensive preprocessing, including partitioning into SQLite databases to enable faster queries during training. Two PyTorch datasets are utilized: one predicts the future color of the center pixel, and the other estimates the time until the next pixel change.

The modeling approach centers on a Vision Transformer (ViT) that combines patch-based inputs for broader context and pixel-level inputs for finer detail. Training involves CrossEntropyLoss for color predictions and MSELoss for timing, with the AdamW optimizer managing updates. While the project supports both single and multi-GPU setups, bottlenecks like SQLite query speeds and architectural refinements are ongoing challenges. This work remains a hands-on exploration of strategies to better understand and predict the canvas's evolving patterns.

---still working on this

Brain Tumor Segmentation

2024

• Python
• U-Net Architecture
• TensorFlow
• PyTorch
• Keras
• Numpy
• Matplotlib
• scikit-learn

In this project, I developed a deep learning model to segment brain tumors from MRI scans, focusing on improving accuracy and reliability. Using a U-Net architecture, I trained the model on multi-modal MRI data (FLAIR, T1, T1ce, T2) to identify tumor regions in both High-Grade Glioma (HGG) and Low-Grade Glioma (LGG) cases. The process involved carefully preprocessing the data, building the U-Net model with features like Batch Normalization and Dropout to enhance performance, and evaluating it using the Dice Coefficient, achieving scores of up to 0.9950 on the validation set.

Ground Truth

Predicted

Ryusen

2024

• Web Dev
• TypeScript
• AI

Ryusen transforms AI conversations into interactive visual experiences through branching tree structures. Each user input generates multiple AI responses, allowing for exploration of diverse viewpoints. Users can customize the number of responses—ranging from one to five—and adjust the creativity level for each branch, shaping the tone and depth of the dialogue.

The interface allows users to easily drag and rearrange nodes to create a personalized conversation map. Once finalized, these maps can be exported as JSON or ASCII trees for sharing or further use.

I plan to add a couple more features to this. what's on mind currently is allowing the user to select the randomness degree of the AI responses.

Protein Embeddings

2024

• C++
• PyTorch
• Sequence Embeddings

This project is a protein sequence analysis tool designed to simulate Multiple Sequence Alignments (MSA) and create meaningful numerical representations (embeddings) of protein sequences using a Transformer-based neural network model.

MemoViz

2024

• Design
• JavaScript
• Threejs

Unofficial spotify wrapped for wrapping images with music.

Calcam

2025

• Design
• JavaScript
• Gsap
• GLSL

A visualization for how calendars could be.

WhoReads

2024

• Python
• Scikit-learn
• Pandas
• Numpy
• Machine Learning

This project is a personalized book recommendation system that uses collaborative filtering to suggest books tailored to a user's reading preferences. Leveraging a massive dataset of approximately 226 million rows of Goodreads user interactions, the system analyzes user reading histories and cross-references them with books that the user has liked. To identify users with similar tastes, cosine similarity is applied, matching the user's reading list with others who have comparable preferences.

Potatoe Disease Detection

2024

• Python
• Keras
• TensorFlow
• Convolutional Neural Networks
• Matplotlib
• FastAPI
• Postman

Trained and built a CNN model with TensorFlow to detect and classify whether a potato leaf is healthy. I also created a FastAPI backend to upload images of potato leaves --I used Postman-- and receive both prediction and confidence levels.