Ali Abul-Hawa, PhD

Data scientist · ML engineer

Applied ML, retrieval, and reliable software systems.

Selected technical work across large-scale NLP, semantic retrieval, document intelligence, computer vision, and reliable software systems.

Electrical engineering background with work spanning machine learning, retrieval, software systems, and control.

Selected work.

Professional work · Information retrieval

Production Semantic Retrieval

MPNet-family semantic retrieval integrated into an existing Ruby and Elasticsearch product, used for both related-skill expansion and live job-title matching.

Compared
9 embedding variants
Evaluated
28,400 relation cases
Production use
Related skills + live job-title search
Skill relations were separated by semantic type, nine embedding variants were evaluated on 28,400 relation cases, and the selected representation powered both related-skill generation and live semantic ranking against precalculated job-title embeddings.
Case study →
Completed team project · Computer vision · MVTec ADIndividual contribution documented

Industrial Visual Anomaly Detection

Industrial anomaly detection and localization across all 15 MVTec AD categories using PatchCore, convolutional autoencoders, ResNet representation experiments, and DINOv2/DINOv3 nearest-neighbour baselines.

DINOv2
0.941 mean image F1
PatchCore
0.929 mean image F1
DINOv3
0.680 mean AUPIMO
My contribution: EDA/reporting, metric and threshold design, autoencoder and ResNet experiments, PatchCore evaluation, fair comparison fixes, and DINOv2/DINOv3 experiments.
Case study →Inspection & localization →
2017-2018 · Professional research · Arabic NLP · Data engineeringCollaborative research

Arabic Social Media NLP at Scale

A large-scale Arabic social-data pipeline covering collection, geolocation, spam removal, normalization, dialect variation, supervised classification, and database-scale preprocessing.

Scale
700M+ raw records
Language
MSA + five dialects
Labeling
~10K expert-labeled samples
Processing hundreds of millions of records pushed preprocessing from per-document and in-memory work toward database aggregation, batching, and disk-backed operations, while the language pipeline had to accommodate MSA and five Arabic dialects.
Case study →Pipeline deep dive →

Independent engineering

Software systems and practical tools.

Independent work spanning backup reliability, language learning, constraint-based generation, and focused desktop utilities. HomeDrive is the most substantial systems project.

Independent engineering · Android + WindowsPrivate repository

HomeDrive

Android media discovery and local state are paired with a Windows receiver that verifies each transfer independently before committing it to retained backup state. Physical testing covers forced interruption, resumable continuation, stale-connection recovery, destination health, and deletion safety.

MediaStore→authenticated transfer→resumable .part→SHA-256 verification→atomic commit
System case study →
Resumable
interrupted uploads continue from retained partial state
Verified
length and SHA-256 checked before final promotion
Fail-closed
contradictory state blocks destructive actions
Full-stack learning system

German Master

Lexeme-centric practice, deterministic task generation, Postgres, enrichment, offline PWA behavior, queued synchronization, and shared history.

Inspect project →

Earlier research

Dynamic systems, nonlinear control, and energy modeling.

PhD work modeled PEM fuel-cell/load interaction in Simulink and developed nonlinear power/voltage control across changing demand, including protection and recovery in concentration-loss conditions. Later work included photovoltaic sizing and life-cycle economic analysis.

Research & engineering →
Historical ML100-class food recognition

VGG16 transfer learning on UECFOOD100 with bottleneck features, fine-tuning, augmentation, and a phone-to-model demo.

Open historical case study →

Contact

Get in touch.

For professional, project, or technical enquiries, use the address below.

qortxai@outlook.com

Open email app