Mehdi Saffar

This personal project is a fluid simulator application written in WebGPU. It utilizes the Smoothed Particle Hydrodynamics (SPH) technique for simulation. The front end is developed using SvelteKit, Flowbite, and TailwindCSS and is bundled with Vite. The simulation and rendering are entirely done in WebGPU, providing a fast and efficient fluid simulation experience.

An online app where users can upload a list of academic articles and visualize the social network of their authors. Users can visualize interesting relations between them with the help of a directed graph and social network analysis (SNA) metrics. The app uses a novel approach where all analysis is performed locally on the front-end using WebAssembly and Web Workers. This was part of a university term project.

An eCommerce app that connects vendors and customers on a single platform. As part of a year-long software engineering course, I led the entire team of 13 people and the front-end sub-team of four people. We used the Ant Design library with React to create a slick-looking website with all the required functionality.

The website includes product listing and filtering, reviewing, checking out a shopping cart, tracking orders, and chatting with vendors. The project was first designed by collecting requirements via continuous communication with the client, making various diagrams that clarified what the system was about and how it worked, and mockups that showed how various use cases would look on the website.

The website was containerized using Docker and deployed on AWS EC2 instances. I set up the CI/CD pipeline using GitHub Actions, which saved a lot of time and human errors during the delicate deployment process. Finally, we collaborated to write a detailed report that explained the entire process behind building and publishing the project and the tools that helped us achieve this goal as a team.

An email classifier, built in Python, that tags emails as spam or not spam based on the Multinomial Naive Bayes (MNB) technique. The classifier is sent a dataset of labeled emails, and after training, it outputs a model that can be used to classify. It can be configured to consider all words used in emails or use the mutual information (MI) metric to choose only the top k=100 discriminating words. This has the benefit of reducing the size of the model and the overfitting on average.

I developed this project for an information retrieval course, and the report can be found on GitHub under report/report.pdf.