Master's student in Computer Science at the American University of Beirut. My background is in machine learning, NLP, and GPU/high-performance computing, and through my master's I'm moving toward ML systems (SysML), edge / on-device ML, and ML security — how models can be run efficiently, close to where data is produced, and safely.
- ML systems: efficient training and inference, especially under limited compute
- Edge and on-device machine learning
- Security for machine learning systems
- Natural language processing, especially Arabic NLP and post-OCR / text correction
- Fine-tuning and evaluating language models with limited resources (LoRA / QLoRA)
- GPU computing and high-performance parallel algorithms
- Applied machine learning on structured and biomedical data
-
Arabic Post-OCR Correction — Model Benchmark Benchmarked five model families (AraFix, AraT5, Qwen, AraBERT, ByT5) on correcting errors in OCR'd Arabic text, on a shared held-out set. A fine-tuned AraFix model reached a character error rate of 0.065, clearly ahead of the rest.
-
GPU Sparse Triangular Solve Four CUDA kernels for the sparse triangular solve, each optimizing the last through level-scheduling and caching, reaching a ~132× geometric-mean speedup over the CPU reference — top of the class benchmark for that milestone.
-
Fall Detection from Human Pose Real-time fall detection from video: OpenPose extracts body keypoints and a Random Forest classifies fall vs. no-fall (~94% accuracy) — a lightweight, edge-friendly design.
-
Breast Cancer Type Classification Multi-class classification on the METABRIC clinical dataset with XGBoost (GPU-trained, tuned) and a neural-network baseline.
-
Molecule Toxicity Classification Toxicity prediction on a small, imbalanced set of molecular descriptors, focused on feature selection and handling class imbalance with SMOTE.
- Languages: Python, C, CUDA, JavaScript
- ML / NLP: PyTorch, Hugging Face Transformers, XGBoost, scikit-learn, TensorFlow/Keras, QLoRA / Unsloth
- Infrastructure: AWS (Bedrock, Textract, RDS, ECS), PostgreSQL / pgvector, MySQL, Flask, Docker
- MSc in Computer Science, American University of Beirut — in progress
- BSc in Computer Science, American University of Beirut — Dean's Honor List (Fall 2024–2025)
- Email: aad40@mail.aub.edu
- LinkedIn: Ali Daher