Mahmoud Abusaqer

IUG Timetabling Using Evolutionary Strategy With Shotgun Hill-Climbing Algorithms

Official implementation of RoBERTa-OTA for multiclass hate speech detection. Combines RoBERTa with ontology-guided graph neural networks for fine-grained demographic targeting classification. Accepted at CSCI'25. Achieves 96.06% F1-score on 5-class hate speech detection.

Official implementation of BERT-OTA: Enhancing Hate Speech Detection with Ontology-Guided Transformer Attention. Combines BERT embeddings with ontology-guided graph attention networks for improved hate speech detection. Published in IEEE Access.

Official implementation comparing parameter-efficient LoRA fine-tuning (Llama, Phi, Qwen families) with zero-shot API classification (o3, DeepSeek, Claude) for abusive language detection. Includes training pipelines and evaluation code. Published in JVLC

A simple GUI Python application that simulate a Goal-Based AI Agent that goes to a target location

Autonomous AI agent that handles university course registration queries using agentic architecture and Claude AI. Built as a proof of concept for AI-powered institutional support systems.