Dr. Daryna
Dementieva

I am a postdoctoral researcher in the Data Science and Analytics Group at the Technical University of Munich. I do research on the Democratic and Responsible AI and Language Technologies: transparent and resource-efficient language technologies built for communities, public trust, and societal impact.

My work spans AI for societal domains, including proactive mitigation of harmful content (TextDetox, HatePRISM, NLP for Social Good); creation of massively multilingual benchmarks and support for low-resource languages (BRIGHTER, UkrDetect); efficient LLMs under resource constraints (WMT shared task); and explainable AI decisions (explainable detoxification, self-explaining hate speech detection). Currently, I focus on rigorous evaluation of reasoning and of LLM applications in high-risk societal domains.

I hold a PhD from Skoltech and have authored 50+ peer-reviewed papers, recognised with the ACL 2025 Best Resource Paper Award and by UNESCO. I am a strong supporter of open source (3M+ downloads of our models and datasets) and of communities advocating AI for Good: I organise the NLP for Positive Impact workshop, the TextDetox and LLMs with Limited Resources shared tasks, and Munich🥨NLP meetups, and mentor in the ACL Mentorship and AI Explorers programmes.