Synthetic Data Generation for Multi-Hop Logical Reasoning
2025
Multi-agent frameworks and SAT-grounded pipelines to generate robust evaluation datasets for complex reasoning in LLMs.
A showcase of research prototypes, production AI systems, and open-source tools spanning Sequential Decision Making, Human-AI Interaction, and Hybrid-Agent Systems.
Automated Planning and Scheduling, Markov Decision Processes, Reinforcement Learning, Search Algorithms.
Cognitive Modeling, Human-Policy Generation, Human-Aware Teaming.
Large Language Models + Automated Planners (TRIP-PAL and GenPlanX), Multi-hop Synthetic Data for Logical Reasoning Agents, Diffusion Models with Constrained Optimization, Secure agents in Pyodide.
2025
Multi-agent frameworks and SAT-grounded pipelines to generate robust evaluation datasets for complex reasoning in LLMs.
2024
Combining Large Language Models (LLMs) and automated planners to create verifiable, constraint-satisfying hybrid agent systems for complex workflows.
2024
Mitigating downside risks in actionable algorithmic recourse policies under stochastic execution (AAAI-24 Oral).
2023
Generating physically/financially constrained synthetic time-series and multi-modal semantic retrieval (NeurIPS 2023 & ICAIF Best Paper).
2023
Open-source Python toolkit enabling automated planners and deep RL agents to benchmark interchangeably on complex relational domains.
2022
Scheduling algorithms accounting for stochastic task durations and human context-switching costs, deployed into production (ICAPS 2022 & US Patent).
2021
A hybrid symbolic planning and monitoring system that achieved top performance in the DARPA SAIL-ON novelty adaptation program.
2016
Embedded communication architectures and occupancy-responsive control systems for commercial and residential automation (Granted US Patents).