I am an AI Research Lead at J.P. Morgan Chase (AI Research Team). I completed my Computer Science Ph.D. in 2022 at Arizona State University in the Yochan Lab under Prof. Subbarao (Rao) Kambhampati.
My work focuses on Sequential Decision Making, Human-AI Interaction, and Hybrid-Agent Systems that unify generative models (LLMs, diffusion) with classical reasoning (automated planning, formal constraints, and optimization).
🎯 Core Research Focus
Sequential Decision Making
Developing rigorous models and algorithms for multi-stage decision problems under uncertainty, stochastic execution, and temporal makespan constraints. Spanning Automated Planning & Scheduling (ICAPS), Markov Decision Processes (MDPs), Reinforcement Learning (RL), and combinatorial optimization.
Human-AI Interaction & Teaming
Integrating insights from cognitive science and psychology into AI algorithms. Modeling human cognitive limitations, perceptual state aliasing, context/task-switching overheads, and intent predictability into policy synthesis, explainable AI (XAI), and safer algorithmic recourse (SafeAR).
Hybrid-Agent Systems (Generative + Classical AI)
Bridging the linguistic fluency of Large Language Models (LLMs) and diffusion models with the strict correctness guarantees, world models, and search capabilities of symbolic automated planners and optimizers. Architecting reliable multi-agent systems, multi-hop reasoning synthetic data generation, and secure browser-based coding agents.
🔬 Featured Publications
The Curious Case of Planning for Unreliable Agents: Challenges and Opportunities in Orchestrating Generative AI Agents
Workshop on Planning in the Era of LLMs at the International Conference on Automated Planning and Scheduling (ICAPS-26)
Investigates orchestrating swarms of stochastic, unreliable generative AI agents using symbolic automated planners with formal recovery and dynamic replanning.
Generating Domain Specific Natural Language SAT Reasoning Datasets
Workshop on Efficient Reasoning at the Conference on Neural Information Processing Systems (NeurIPS 2025)
Introduces programmatic synthesis pipelines to translate Boolean satisfiability (SAT) problem instances into natural language narratives for benchmarking multi-hop logical deduction in LLMs.
SafeAR: Towards Safer Algorithmic Recourse by Risk-Aware Policies
Proceedings of the AAAI Conference on Artificial Intelligence (AAAI-24)
Formulates risk-aware recourse policies under stochastic execution and uncertain outcomes via Conditional Value-at-Risk (CVaR) objectives.
Multi-Modal Financial Time-Series Retrieval Through Latent Space Projections
Proceedings of the 4th ACM International Conference on AI in Finance (ICAIF 2023)
Aligns financial text, unstructured news, and multivariate market continuous time series into a joint latent representation for semantic cross-modal retrieval.
On the Constrained Time-Series Generation Problem
Advances in Neural Information Processing Systems (NeurIPS 2023)
Marries score-based diffusion models with projected gradient methods to guarantee strict non-convex domain and safety constraint satisfaction in synthetic time-series.
TRIP-PAL: Travel Planning with Guarantees by Combining Large Language Models and Automated Planners
arXiv preprint arXiv:2406.10196
One of the first papers to combines LLMs for rich preference extraction with symbolic automated planners to generated valid, constraint-satisfying travel itineraries.