I’m a PhD student in Computer Science at the Ubiquitous Knowledge Processing (UKP) Lab, TU Darmstadt in Germany, supervised by Prof. Iryna Gurevych.
I work on NLP and LLMs, with a focus on reliable knowledge access and LLM safety. I’m interested in how systems retrieve and use evidence, and what happens when important evidence is missing.
My recent work includes CORE-T, a framework for retrieving coherent sets of tables for text-to-SQL. I’m also investigating the robustness of retrieval-augmented and agentic systems to incomplete evidence.
Previously, I was an AI researcher in DFKI’s Educational Technology Lab in Berlin, supervised by Dr. Miloš Kravčík. I worked on educational chatbots, retrieval-augmented mentoring, and adaptive dialogue.
I completed my MSc in Computer Science at Saarland University in Germany. My master’s thesis on cross-domain neural entity linking was supervised by Prof. Dietrich Klakow at Saarland University, in collaboration with Bosch Center for AI (BCAI), where I worked with Dr. Heike Adel, Dr. Mohamed H. Gad-Elrab, Dragan Milchevski, and Dr. Jannik Strötgen.
CORE-T accepted to EMNLP 2026 (Main).
Released the CORE-T preprint on coherent multi-table retrieval for text-to-SQL.
Started tutoring Introduction to LLMs at TU Darmstadt for winter semester 2025–26.
Retrieval-Augmented Chatbots for Scalable Educational Support in Higher Education published in the GenAI-LA @ LAK 2025 proceedings.
Presented our educational chatbot work at GenAI-LA @ LAK 2025.
Started my PhD at UKP Lab, TU Darmstadt.
Scalable Mentoring Support with a Large Language Model Chatbot published at ECTEL; received the Second Best Demo Paper award.
Our work on LLM-based learning support in educational sciences appeared at DELFI.
Our adaptive dialogue management paper published in ACM UMAP Adjunct ’24.
My ongoing SafeLLMs research studies strategic missingness: how omitted evidence can affect the conclusions of LLM-based systems. I’m exploring evidence coverage, retrieval, and evaluation in retrieval-augmented generation and agentic fact-checking.
Ongoing researchHow can a system find not just individually relevant tables, but a set that works together? In CORE-T, we study training-free, join-coherent table retrieval for open-book text-to-SQL over large, heterogeneous collections.
CORE-T publication ↓My earlier work explored LLM-based mentoring and dialogue systems that adapt to their users, alongside entity linking across domains.
For the full list of publications, visit Google Scholar.
Training-free retrieval of compact, join-coherent table sets for answering questions across heterogeneous table collections.
Retrieval-augmented chatbots grounded in course materials to provide scalable, personalized support for university students.
An LLM-based chatbot that supports mentoring conversations and helps students engage with course content.
Second Best Demo Paper
Integrating an LLM-based learning companion into teacher education to support students in educational sciences.
Adapting telephone-assistant dialogue to users’ emotional states and demographic characteristics with LLMs.
Combining general- and domain-specific training data to improve entity linking across knowledge graphs without sacrificing general-domain performance.
Darmstadt, Germany
UKP Lab · Supervised by Iryna Gurevych
Berlin, Germany
Educational Technology Lab
Renningen, Germany
NLP & Semantic Reasoning
Renningen, Germany
In collaboration with Saarland University
Saarbrücken, Germany
Database & Information Systems
Luxembourg, Luxembourg
Fulfillment Acceleration · Luxembourg
Saarbrücken, Germany
Saarland University · Overall GPA: 1.40 / 1.00
Thesis: Cross-Domain Neural Entity Linking
Supervisor: Prof. Dietrich Klakow
Alexandria, Egypt
Alexandria University · Overall GPA: 3.96 / 4.00 · First Class Honours
Thesis: Car License Plate Arabic Numbers & Letters Recognition
Supervisor: Dr. Marwan Torki
For Scalable Mentoring Support with a Large Language Model Chatbot.
Recognized for outstanding academic performance in Computer & Communication Engineering.
Implemented a 135M-parameter language model and fine-tuned it for grammatical error correction.
PyTorch · Hugging Face · TRL
Compared multilingual embeddings, TF-IDF with Multinomial Naive Bayes, and fine-tuned transformers for language identification across 20 languages.
Multilingual embeddings · Scikit-learn · Transformers
TU Darmstadt · Darmstadt, Germany
NLP foundations, LLM architectures and training, prompting, and fine-tuning. Practical topics include information retrieval, dense retrieval, rerankers, retrieval-augmented generation, agentic approaches, and hands-on model evaluation.
TU Darmstadt · Darmstadt, Germany
Remote · Darmstadt, Germany
For research questions or to discuss related work, you can reach me at hassanmahmoudsd@gmail.com.
Email
GitHub
LinkedIn
Scholar
Résumé
