Shuhaib Mehri

prof_pic.jpg

Hi 👋! I am a PhD Student at the University of Illinois Urbana-Champaign, advised by Prof. Dilek Hakkani-Tur.

I work on user simulation. Some recent highlights:

  • Introduce a method for measuring the distributional gap between real and simulated user behaviors, evaluate 24 LLM-based user simulators, and show that combining behaviorally complementary ones narrows the gap [paper].
  • First to show that user simulators in widely used benchmarks (e.g., Ï„-Bench) do not reliably follow their assigned goals. We introduce user goal state tracking and train goal-aligned simulators [tweet] [paper].
  • Built a benchmark for multi-session collaboration, and use simulator behavior as an RL training signal for improving agent memory [tweet] [paper].

I am super excited to simulate users at Microsoft Research this summer. Previously, I have interned at IBM Research, and Amazon (x3).

news

Mar 24, 2026 Very excited to virtually attend EACL 2026! I will be giving an oral presentation on our work on reference-level feedback. I’ll also serve as a session chair for Tuning & LLM Behaviour!
Feb 13, 2026 Excited to give a talk at Google Research on user-centered conversational agents! I will cover (1) LLMs that reliably simulate users during multi-turn conversations, and (2) agents that continuously learn user preferences & improve collaboration across multiple sessions.
Feb 03, 2026 I am presenting our recent preprint MultiSessionCollab: Learning User Preferences with Memory to Improve Long-Term Collaboration at the Assistive and Inclusive Technology workshop today.
Dec 11, 2025 Came full circle with a visit to the UBC NLP Group (where my research journey began!) and gave a talk on goal alignment for LLM-based user simulators.
Sep 18, 2025 Our work on user simulator goal alignment has been accepted to TACL!
Jan 08, 2024 I was awarded the Mitacs Accelerate Fellowship for Master’s and PhD Students as an undergrad. Really excited to start working with iClinic Systems Inc.!

selected publications

  1. Measuring and Mitigating the Distributional Gap Between Real and Simulated User Behaviors
    Shuhaib Mehri, Philippe Laban, Sumuk Shashidhar, Marwa Abdulhai, Sergey Levine, Michel Galley, and Dilek Hakkani-Tür
    Preprint, 2026
  2. MultiSessionCollab: Learning User Preferences with Memory to Improve Long-Term Collaboration
    Shuhaib Mehri, Priyanka Kargupta, Tal August, and Dilek Hakkani-Tür
    Preprint, 2026
  3. oral
    Goal Alignment in LLM-Based User Simulators for Conversational AI
    Shuhaib Mehri, Xiaocheng Yang, Takyoung Kim, Gokhan Tur, Shikib Mehri, and Dilek Hakkani-Tür
    TACL 2025 (Oral at ACL 2026), 2025
  4. oral
    Beyond Sample-Level Feedback: Using Reference-Level Feedback to Guide Data Synthesis
    Shuhaib Mehri, Xiusi Chen, Heng Ji, and Dilek Hakkani-Tür
    EACL 2026 (Oral), 2025