Engineer & Researcher — Computer Vision · Deep Learning · Document AI

Engineering Degree + Master's, Artificial Intelligence & Data Science — École Supérieure en Informatique (ESI), Sidi Bel Abbès, Algeria (Jul 2026).

Open to PhD positions and permanent (CDI) roles, in computer vision, document AI, and applied deep learning.

(Both papers are currently in preparation with co-authors. Drafts are available directly — see links above or contact me — while the arXiv submission is finalized.)

multimodal LLMs OCR alignment weak supervision document layout analysis

Built a reference-grounded annotation pipeline (RefLAM) that aligns multimodal-LLM OCR against independently sourced clean transcriptions, and used it to construct AraMS-28k — the largest publicly released line-level dataset of genuine historical Arabic manuscripts (14 books, 3,043 pages, 28,600 lines, three script traditions). Introduced the first publicly released line-level insertion-anchor annotation, recovering non-linear main/margin reading order. Proved a formal correctness guarantee (the Confidence-100 rule) achieving a 75× annotation throughput gain over manual review, with zero errors found on independent audit. 📦 Repository · 📄 See Publications above

PyTorch Transformers OpenCV video understanding

Led a 5-person team extending the TEVAD baseline for video anomaly detection through architectural refinements and improved multi-modal fusion, evaluated on UCF-Crime subsets. 📦 Repository · 📊 AUC 98.6% → 99.2%, 11× faster convergence · 📄 Technical report

GPT FastAPI React NL-to-SQL

Built a natural-language interface for querying structured datasets conversationally, via a multi-stage pipeline (intent classification → SQL generation → validation → interpretation), with automated data profiling to flag anomalies and inconsistent distributions. 🔗 Live system · 📊 Validated on 100+ real-world datasets

PyTorch ResNet MobileNet transfer learning

Classified COVID-19, pneumonia, and normal chest X-rays using transfer learning with data augmentation and class balancing, deployed as an interactive demo. 🔗 Live demo · 📦 Repository · 📊 F1 0.93, ROC-AUC 90.22%

Document & Image Analysis: Handwritten Text Recognition (HTR) · OCR-reference alignment · layout analysis · image restoration · Kraken · HATFormer Deep Learning & Computer Vision: PyTorch · TensorFlow · OpenCV · Vision Transformers · Multimodal LLMs · generative models NLP: Transformers · BERT variants · CLIP · GPT models · Hugging Face ecosystem Engineering: Python · SQL · Git · Docker · Linux · Weights & Biases Languages: Arabic (Native) · French (B2) · English (Professional)

_{📫 Reach out about research collaboration, PhD opportunities, or roles in document AI / computer vision.}