Shuhaib Mehri

Hi 👋! I am a PhD Student at the University of Illinois Urbana-Champaign, advised by Prof. Dilek Hakkani-Tur. I try to build AI that is useful to people in the real world, and my current research focuses on user simulation, which I believe to be an important part of the puzzle. Some recent highlights:

• Identified user representations as a bottleneck in user simulation, and trained UserRepGen-4B to generate representations that lead to more faithful simulation [paper].
• Measured the distributional gap between real and simulated user behaviors [tweet] [paper].
• Built MultiSessionCollab, a benchmark for multi-session human–AI collaboration, and leveraged user simulators to train collaborative agents that use memory more effectively [tweet] [paper].
• Showed that user simulators in interactive benchmarks (e.g., τ-Bench) do not actually follow their assigned goals. Introduced user goal state tracking and trained goal-aligned user simulators [tweet] [paper].

I interned at Microsoft Research, where I had the wonderful opportunity to work with Michel Galley from the Deep Learning Group and Philippe Laban from the AI Interaction and Learning Group. Before that, I also interned at IBM Research, and Amazon (x3).