Home
About me

Mert Cemri

alt text 

Mert Cemri Ph.D. Student at UC Berkeley Berkeley, CA, United States

Email: cemri@berkeley.edu

[Google Scholar] [Curriculum Vitae]

I am a Ph.D. student in the Berkeley AI Research (BAIR) and Sky Computing Labs at UC Berkeley, advised by Ion Stoica, Kannan Ramchandran, and Alex Dimakis.

I am also a part of the Berkeley Laboratory for Information and System Sciences (BLISS) and serve as an organizer of the BLISS Seminar. Feel free to contact me if you want to participate in our seminar!

My research focuses on efficient machine learning and agentic systems, including multi-agent large language model (LLM) systems, test-time scaling, and discrete generative models.

Selected publications

  • Why Do Multi-Agent LLM Systems Fail?
    M. Cemri, M. Pan, S. Yang, L. Agrawal, B. Chopra, R. Tiwari, K. Keutzer, A. Parameswaran, D. Klein, K. Ramchandran, M. Zaharia, J. E. Gonzalez, I. Stoica. Advances in Neural Information Processing Systems (NeurIPS), Datasets and Benchmarks Track, 2025.

  • SPECS: Faster Test-Time Scaling through Speculative Drafts
    M. Cemri, N. Rajaraman, R. Tiwari, X. Liu, K. Keutzer, I. Stoica, K. Ramchandran, A. Beirami, Z. Sun. ES-FoMo III: 3rd Workshop on Efficient Systems for Foundation Models (ICML), 2025.

  • DigiRL: Training In-The-Wild Device-Control Agents with Autonomous Reinforcement Learning
    H. Bai, Y. Zhou, M. Cemri, J. Pan, A. Suhr, S. Levine, A. Kumar. Advances in Neural Information Processing Systems (NeurIPS), Main Track, 2024.

  • Asynchronous Social Learning
    M. Cemri, V. Bordignon, M. Kayaalp, V. Shumovskaia, A. H. Sayed. IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2023.

  • Discovering Influencers in Opinion Formation over Social Graphs
    V. Shumovskaia, M. Kayaalp, M. Cemri, A. H. Sayed. IEEE Open Journal of Signal Processing, 2023.

For a complete list of publications, please see my Google Scholar profile or my CV.

Experience

I have interned as a machine learning engineer at Apple, working on agentic post-training for foundation models, and previously conducted research at EPFL with Ali H. Sayed on distributed optimization and social learning, and at Bilkent University with Tolga Çukur and Aykut Koç on neural methods for natural language and graph-structured data.

Contact

The best way to reach me is by email at cemri@berkeley.edu