a portfolio by

EMRAAN YUSUF

SOFTWARE ENGINEER · ML RESEARCHER

DALLAS, TX

SYNOPSIS [About]

GenreML / FULL-STACK

StudioUT DALLAS

RuntimeYEAR 3

I'm Emraan, a Computer Science student at the University of Texas at Dallas with a focus on machine learning and research.

I got into technology at an early age through a fascination with how the tools I use every day work, which led me to study Computer Science at the University of Texas at Dallas. During my time at UTD, I’ve developed a strong passion for Machine Learning, especially research. Today, I spend most of my time learning and building; working on machine learning research projects, keeping up with research developments, creating full-stack applications and attending hackathons.

Outside of coding, I’m usually in the gym, at my mosque, or watching movies. Film is something I've always been passionate about, inspiring the film-themed design of this portfolio, from the reel countdown to the perforated borders and other cinematic elements throughout!

[ DOWNLOAD RESUME ]

·· TAIL OF REEL 01 ··|·· HEAD OF REEL 02 ··

FILMOGRAPHY [Projects]

Conflict Coordinate

Real-time conflict event mapping using conflict tracking sources and ML classification.

DC Daddy

Centralizing and simplifying complex data-center operations into one intuitive platform.

QuranScope

Semantic search across Quranic text with support for AI-powered summaries and questions.

Hypertrophy Tracker

Science-based workout logging and progressive overload tracking for hypertrophy.

BJJ Simulator

Interactive Brazilian Jiu-Jitsu position and submission viewer.

Caketopia

Cake stacking game in Greenfoot with physics elements implemented.

Assembly Multiplication Game

Multiplication card game built entirely in MIPS Assembly.

·· TAIL OF REEL 02 ··|·· HEAD OF REEL 03 ··

PRODUCTION CREDITS [Experience]

Led a team of 12 researchers producing 200+ weekly annotations to build datasets for detecting logical fallacies in political memes
Created comprehensive annotation guidelines to standardize how logical fallacies are identified across the team
Used Python to analyze annotation results and compare them with model predictions
Helped develop a structured fallacy dataset to be used by anyone conducting Natural Language Processing research once published
Contributed to a research project being submitted to the Association for Computational Linguistics (ACL)

Designed and evaluated visual prompt injection attacks on multimodal AI systems to study how hidden instructions in images affect model behavior
Built a testing pipeline using Python, PyTorch, and NumPy to generate attacks, run experiments, and measure attack success rates across multiple models
Worked with open-source multimodal models from Hugging Face to benchmark vulnerabilities in models such as Gemma, Llama, and Qwen
Benchmarked attack success rates at 86-91%, then designed defense methods to reduce the attack success rate by 99%, effectively completely mitingating the problem in tests
Maintained accuracies and inference times that remained close to the baseline after applying defenses

GPA: 3.6
Relevant coursework: Artificial Intelligence, Machine Learning, Databases, Networks, Data Structures and Algorithms, Software Engineering