Headshot

Hi! I am currently a research scientist on the prediction and planning team at Waymo.

I received my Ph.D. from the EECS department at UC Berkeley, where I was advised by Sanjit Seshia and supported by the NDSEG Fellowship. My thesis broadly focused on the development of neuro-symbolic techniques to improve the safety, reliability, and consistency of robotic and autonomous behavior. Please see my research for more details.

Before coming to Berkeley, I completed my undergraduate and Master's training at Rice University, where I worked on projects related to interpretable machine learning via program synthesis and formal methods. I was advised by Swarat Chaudhuri, and also collaborated on projects with Ankit Patel and Richard Baraniuk.

In my free time, I enjoy doing a lot of things, including but not limited to: playing tennis, reading novels, inventing new recipes, and repeatedly fixing and breaking my jumpshot. I'm originally from Cleveland, Ohio, and I always make sure to watch my beloved Cavaliers break my heart year after year.

You can find me on X or on Google Scholar. The best way to reach me is by sending me an email at ameesh@berkeley.edu.

Recent News (see all)

03/2026
I started at Waymo! Looking forward to building something great here.
02/2026
Our work on learning to dynamically improve Vision-Language-Action models during inference time has been accepted to ICRA 2026!
01/2026
I'm once again co-organizing the VerifAI ICLR Workshop on bridging semi-formal verification and foundation models!
12/2025
I gave my dissertation talk! Thank you to everyone at Berkeley who was part of this journey :)
05/2025
I attended AAMAS in Detroit to present my work on multi-agent RL for symbolic tasks!

Publications

Refereed Publications

Learning Affordances at Inference-Time
for Vision-Language-Action Models
Robust and Diverse Multi-Agent Learning via Rational Policy Gradient
Learning Symbolic Task Decompositions for Multi-agent Teams
A. Shah, N. Lauffer, T. Chen, N. Pitta and S. A. Seshia
Learning Formal Specifications from Membership and Preference Queries
LTL-Constrained Policy Optimization with Cycle Experience Replay
Who Needs to Know? Minimal Knowledge for Optimal Coordination
Learning Deterministic Finite Automata Decompositions
from Examples and Demonstrations
Modeling and Influencing Human Attentiveness
in Autonomy-to-Human Perception Hand-offs
Learning Differentiable Programs with Admissible Neural Heuristics
Representing Formal Languages: A Comparison Between Finite
Automata and Recurrent Neural Networks
J. J. Michalenko, A. Shah, A. Verma, S. Chaudhuri and A. B. Patel

Technical Reports and Preprints

arXiv '23
arXiv '23
Specification-Guided Data Aggregation for Semantically
Aware Imitation Learning
arXiv '20
arXiv '20
Demonstration Informed Specification Search
M. Vazquez-Chanlatte, A. Shah, G. Lederman and S. A. Seshia
Rice Thesis '20
Rice Thesis '20
Differentiable Program Learning with an Admissible Neural Heuristic
A. Shah