Generative AI that recommends news: personal, but still diverse, transparent and good for public debate.
Doctoral researcher & UZH.ai Fellow, University of Zurich · Research: News Recommender Systems, Generative AI, Computational Social Science, Communication Science, NLP, Collective Intelligence
I am a doctoral researcher in the DDIS group at the University of Zurich, a UZH.ai Fellow and a scholar in the DSI Excellence Program. My PhD asks how generative AI can personalize news without narrowing what people see, and I test the answers with real readers in field studies on Informfully.
I came to AI the long way around: business administration in St. Gallen, private equity, applied economics research at ETH Zurich, then an MSc in AI at UZH. That path is why I work where computer science meets communication and social science. Out of personal interest, I also explore collective intelligence: whether swarms of foundation models reason better together than alone.
Noah Mamié and Laurin van den Bergh. The 9th BlackboxNLP Workshop at EMNLP 2026
Natural-language user profiles enable people to interact with and customize recommendations by providing them with a description of their preferences that enables direct intervention and meaningful steering of the recommender system.
Noah Mamié and Susie Xi Rao. The Sixteenth International Conference on Swarm Intelligence 2025
Foundation model swarms mimicking biological flocking behaviors significantly outperform individual LLMs in complex logical reasoning.
Eric Egli, Noah Mamié, Mathias Müller and Eyal Liron Dolev. SwissText Analytics Conference 2023 · Best Poster Award
Automated stance detection across Swiss official voting booklets reveals subtle linguistic slant across German, French, and Italian.
An end-to-end research platform for conducting empirical field studies. Informfully comprises native mobile applications for iOS and Android delivering personalized text, audio, and video content with behavioral logging and in-app surveys to facilitate rigorous research on recommender systems.
Multilingual stance detection system analyzing official federal voting booklets across German, French, and Italian with fine-tuned M-BERT models to empirically monitor neutrality in Swiss direct democracy. Awarded the SwissText 2023 Best Poster Award .
An evolutionary multi-agent optimization framework orchestrating swarms of diverse foundation models (LLMs) to unlock collective reasoning and emergent intelligence, presented at the 16th International Conference on Swarm Intelligence (ICSI 2025).
High-performance reinforcement learning implementations experimenting with deep Q-networks, policy optimization, and representation learning across classic Atari 2600 benchmark environments.