Walid Aransa | Personal Website

Research Interests

o Large Language Models (LLMs), Generative AI, and Agentic AI Workflows.
o Retrieval-Augmented Generation (RAG) and Semantic Search Pipelines.
o LLM Inference Optimization, Model Quantization, and Personal GPU Infrastructure.
o Multilingual NLP, Hate Speech, and Abusive Content Classification.
o Neural Machine Translation (NMT) and Multimodal Translation.
o Reinforcement Learning (RL), Alignment (RLHF/DPO), and Neural Networks.
o Sentence/Word Continuous Vector Representations (Embeddings).

o Ph.D. LIUM, Université du Maine, Le Mans, France, 2015
o M.Sc. College of Computing & Information Technology, Arab Academy for Science and Technology, 2010
o Diploma Information Technology Institute, Cairo, Egypt, 2001
o B.Sc. Faculty of Science, Suez Canal University, 1997

o Since Sep 2025: Independent AI Research
o Jun 2025 – Sep 2025: Chief Scientist Officer (CSO) at Axcels
o Nov 2024 – Jun 2025: Chief Product Officer (CPO) at Axcels
o Sep 2024 – Nov 2024: Founder in Residence at Antler
o Dec 2020 – Aug 2024: Senior Research Scientist NLP/AI (Consultant) at Huawei R&D, Finland
o Jun 2018 – Nov 2020: Freelance AI Consultant
o Oct 2015 – May 2018: Post-Doc Researcher at LIUM – University of Le Mans, France
o Dec 2011 – Sep 2015: PhD Student/Researcher at LIUM – University of Le Mans, France
o Feb 2014 – May 2014: Visitor Researcher at IBM Watson Research Center in New York, USA
o Mar 2003 – Nov 2011: Team Leader/Advisory Software Engineer at IBM, Egypt
o 2003: CEO and Founder at Orbits Technology, Egypt
o Oct 2001 – Jan 2003: Senior Software Engineer at Ellipsis Digital Systems based in USA, Egypt
o Jan 2001 – Oct 2001: Software Engineer at icomXpress based in USA, Egypt
o Oct 1999 – Mar 2000: Teaching Assistant at Faculty of Computer Science, MSA University, Egypt

o Journal: ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)

o WMT 2016 Multimodal Machine Translation Shared Task (LIUM ranked first in both tasks)
o Continuous Space Language and Translation Model Toolkit – Add the support for auxiliary features and experiment with different types of auxiliary features, since 2014
o BOLT, DARPA 2011-2014, Broad Agency Operational Language Translation with IBM TJ Watson and other research labs in the USA and Europe.
o NIST 2012 Open Machine Translation Evaluation (OpenMT12)
o Extend to new language pairs and support more features for DELiC4MT, Machine Translation Marathon 2012, September 3-8, Edinburgh, UK
o Using Reinforcement Learning to learn how to walk, applied on half-cheetah: The algorithm used is Augmented Random Search (ARS). Check the video online
o Detect objects in a video using Single Shot MultiBox Detector (SSD) method. Check the video online

Please find the list of my publications here

o First place in three shared tasks and second place in the fourth task at HASOC-FIRE 2023 (team FiRC-NLP)
o Top rankings at SemEval 2023 Task 10 (Explainable Online Sexism Detection): 2nd in Task A, 7th in Task B, 4th in Task C (team FiRC-NLP)
o First place in Task 1 (Multimodal Machine Translation) of WMT 2017 Shared Task (with LIUM team)
o First place in both tasks of WMT 2016 Multimodal Machine Translation Shared Task (with LIUM team) - see official results
o Best paper presentation award at CICLING 2016
o ACL Member