Abdelbasset Djamai

• •Designing and deploying multilingual NLP and agentic AI systems for industrial and enterprise use cases.
• •Built a multilingual translation pipeline spanning 30+ language directions for complex XML/DITA technical documentation, with terminology-sensitive processing, RAG-augmented retrieval, and a multi-agent quality evaluation framework across seven assessment dimensions.
• •Acted as technical lead for a multilingual agentic AI system for a major Algerian telecom operator, designing and evaluating a Text-to-SQL agent that supports French, Arabic, and English over relational data with automated visualization, structured output verification, cross-lingual grounding, and multilingual safety filtering.

• •Advancing Arabic NLP through research on Arabic-centric and culturally aware (large) language models.
• •NLP applications in computational social science and humanities contexts (polarization, hate speech, politiccal ideology/stance detection, ...), focusing on Arabic and multilingual settings.
• •Engaged in a computational history project utilizing NLP & knowledge graph techniques to analyze low-resource, dialectal Arabic historical texts collected from oral testimonies.
• •Investigated LLM-generated Arabic text detection using computational linguistic analysis for novel AI-authorship attribution techniques.
• •Building and releasing dialectal translation models, uni- and multimodal resources, embedding/re-ranking models, and evaluation benchmarks for Arabic NLP research and applications in collaboration with the community and external partners.

• •Co-led a project on LLM-based Classical/Historical Arabic-to-English machine translation informed by translation studies theory, addressing lexical nuance, cultural & linguistic translation gaps, and evaluation beyond surface fluency.
• •Led a multimodal vision-language project for hate speech and toxicity detection in multilingual social media, with a focus on under-resourced Arabic dialects including Algerian Darja, investigating both zero-shot and multi-agent approaches.

• •Taught an interdisciplinary course on AI concepts for humanities research to English/Humanities students.
• •Integrated NLP and generative AI with ethical perspectives to enhance student understanding of responsible AI development.

• •Proposed a hybrid framework for Arabic abstractive summarization that combines transfer learning and reinforcement learning to address a core research gap.
• •Designed evaluation and reward strategies targeting summary quality and factual consistency, improving robustness across benchmarks.
• •Achieved around 10% BERTScore gains over prior results with stronger generalization and factual accuracy; resulting manuscript is under review at Information Processing and Management (Elsevier).

AbjadNLP 2026 · EACL

Omar Elshehy, Omer Nacar, Abdelbasset Djamai, Muhammed Ragab, Khloud Al Jallad, Mona A. Azim.

SemEval-2026 · ACL

Abdelbasset Djamai, Sahara H. Al-Madi, Norah N. Al-Zaid, Khloud Al Jallad, Mona A. Azim.

Information Processing and Management (Elsevier)

Abdelbasset Djamai, Fateh Bougamouza.

Algerian AI Symposium

Abdelbasset Djamai, Sarah Bendissa. Poster; presented.