Postdoctoral Research Fellow at MBZUAI, UAE; additionally affiliated with the UKP Lab, TU Darmstadt
Postdoctoral and Doctoral (PhD) Researchers in NLP & AI (m/f/d)
The UKP Lab, led by Prof. Dr. Iryna Gurevych, is one of Europe’s leading groups in Natural Language Processing and is recruiting postdoctoral researchers to join a highly international team at the frontier of trustworthy and applied AI. We combine fundamental research with open-source tools used worldwide (e.g., Sentence Transformers, AdapterHub), and are embedded in a rich ecosystem — ELLIS, Konrad Zuse School ELIZA, hessian.AI, and ATHENE.
We are looking for outstanding researchers who want to shape how humans and AI systems work together — and ensure that they do so reliably, safely, and in the service of people.
Priority research directions
We especially welcome applications in the following areas, though strong candidates across UKP’s wider profile are encouraged to apply:
A. Trustworthy & agentic AI
Agent reliability & integrity. Agentic systems (planning + tool use + memory + long-horizon action) are the dominant theme, and their multi-step trajectories create new failure modes — misalignment, tool-induced myopia, trajectory-level errors, MCP/tool attack surfaces. There's an explicit push toward a “science of AI agent reliability.” This is exactly your “agent integrity by design” thesis. Example topics: trajectory-level evaluation and guardrails; verification-first agent architectures; honest failure vs. confident error.
Evaluation science / verification-first AI. Evaluation is now the recognized bottleneck: LLM-as-judge has known blind spots, reference-free and meta-evaluation are growing, and “agents in science are an evaluation problem, not an automation problem.” Directly extends UKP’s GREP / preference-based evaluation work. Example topics: calibrated, expert-preference evaluation; meta-benchmarks; verification that scales when content is cheap but checking is expensive.
Interpretability & robustness for safety. Mechanistic interpretability, red-teaming, and robustness under distribution shift are core to “well-understood, reliable models” (ATHENE fit). Example topics: interpretability methods that feed directly into safety guarantees rather than post-hoc explanation.
B. Human–AI collaboration & AI for science
Human–AI co-construction / HCI for expert work. Human-centered NLP and human–AI interaction are explicitly called in major NLP and AI venue, and UKP’s HAI-Co² framing is ahead of the curve. The “cognitive offloading / drift” risk (producing results without understanding) is a hot HCI+NLP problem. Example topics: mixed-initiative interaction, preference elicitation, scaffolding that preserves human skill rather than replacing it.
AI for science & scholarly NLP. Hypothesis generation, experiment design, related-work generation, and especially peer-review support / the peer-review crisis are very active (including work on hallucinated evidence and the coming flood of AI-accelerated papers). Strong overlap with UKP’s cross-document NLP and InterText. Example topics: trustworthy research agents; reproducibility and provenance for AI-assisted science.
C. NLP for high-stakes domains
Misinformation, provenance & AI-text forensics. Claim verification at scale, content provenance/watermarking, and AI-generated-text and deepfake detection sit at the intersection of your fact-checking strength and the ATHENE cybersecurity mission. Example topics: verification-first fact-checking; detection that's robust to adversarial paraphrase.
Privacy-preserving & health NLP. Private/federated adaptation, clinical and mental-health NLP, and compliance pressure (EU AI Act) make this very timely. Example topics: privacy-preserving LLM adaptation; safe clinical/mental-health agents with human oversight.
Legal & argumentative NLP. Legal reasoning is a central NLP challenge , and computational argumentation extends naturally into AI & democracy / deliberation. Example topics: argument-based fact-checking; structured legal reasoning; deliberative AI for public discourse.
D. Foundations that cut across all of the above
Reasoning, planning & neuro-symbolic methods — faithful “System-2” reasoning, process supervision, tool-use that improves accuracy without degrading reasoning quality.
Efficient, modular & continually-adapting models — parameter-efficient and composable models, small/on-device, and continual learning to match the “learn new skills every three months” point.
Multimodal & multilingual/culturally-aligned models — any-to-any models, modality-balance/bias in VLMs, and multilingual safety/evaluation where most safety work is still English-centric.
Your profile
- A PhD or Master degree (completed or near completion) in Computer Science, Natural Language Processing, Machine Learning, HCI, or a related field
- A strong publication record at top-tier venues (e.g., ACL, NeurIPS, ICML, ICLR, EMNLP, NAACL, CHI)
- Solid programming skills and hands-on experience with modern ML/NLP frameworks
- Ability to drive an independent research agenda and to collaborate across disciplines
- Excellent written and spoken English
- Desirable: experience mentoring students, contributing to open-source software, or securing third-party funding
What we offer
- A vibrant, international research environment with strong infrastructure, compute, and engineering support
- Access to leading networks (ELLIS, ELIZA, hessian.AI, ATHENE) and excellent collaboration opportunities
- Mentorship toward research independence and, where desired, support in career development
- A position in Darmstadt, at the heart of Germany’s AI landscape, with TV-TU E13 remuneration, full-time, for initially 2 years with an optional extension, starting as soon as possible
How to apply
Please submit, as a single PDF:
- A cover letter outlining your research interests and fit with one or more of the priority directions
- A full CV with publication list
- A short research statement (1–2 pages)
- Names and contact details of two referees
Send your application quoting reference to your prioritized topics to https://careers.ukp.informatik.tu-darmstadt.de. Review of applications is continuous.
The Technical University of Darmstadt is an equal-opportunity employer. We actively encourage applications from women and from candidates of all backgrounds. Applicants with disabilities will be given preference where equally qualified.
