Michael Kosmider

Master's student at Georgia Tech, specializing in machine learning.

Contact: Email | GitHub | LinkedIn

About Me

I’m a graduate student (4.0 GPA) at Georgia Tech's Online Master of Science in Computer Science, with a passion for developing deep learning algorithms from the ground up. I enjoy digging into the mathematical foundations and implementation details that make these models work, and I find it rewarding to build them from scratch. In addition to my focus on machine learning, I also have experience with multi-threaded systems programming in C/C++ and some knowledge of web development, which support my broader interests in software.

Hard Skills, Concepts, and Projects

Python, NumPy, PyTorch, Hugging Face, C/C++, JavaScript, HTML, CSS.

Deep learning fundamentals, NLP, computer vision, strong mathematical background, multithreaded programming and OOP design.

Image Captioning

I implemented and trained a ViT + decoder image captioner from scratch on the MS COCO dataset, achieving well structured English descriptions like "a herd of sheep standing on top of a lush green field ." I used a masked autoencoder to pre-train the ViT on unlabeled images, and used gradual encoder layer unfreezing for full captioner training. A writeup explains the methods in detail.

View Project

Deep Learning Framework

A full-fledged deep learning framework implemented from scratch in NumPy, paired with a tutorial and documentation. The framework contains standard modules (such as Linear and Convolution layers), functions and optimizers built on top of an autograd system. I successfully trained a CNN using my own framework, achieving over 80% on the CIFAR 10 dataset.

View Project

Transformer Tutorial

An in-depth tutorial on the theory and implementation of transformers following "Attention is all you need". I showed how to build every transformer component from scratch, implemented beam search with KV caching, and demostrated how to train a transformer for the task of machine translation using the multi30k dataset. The transformer components are on PyPi.

View Project

Animations and Drawing App

Two fun, less serious side projects demonstrating proficiency in vanilla HTML, CSS and JavaScript. The first is a set of customizable animations, and the second is a minimal drawing application, where users can select from various shapes, colors, and brush sizes.

View Animations View Drawing app

Education

Master of Science

Georgia Institute of Technology

2024–2026 (expected)

Computer Science with a specialization in Machine Learning. GPA: 4.0

Bachelor of Science

McGill University

Degree Granted in 2023

Major in Mathematics and Computer Science