Reinforcement Learning · LLM Agents · Generalization

Building agents that learn, reason, and generalize.

I study reinforcement learning and LLM-based agents that can learn from experience, reason over complex tasks, and generalize beyond the settings in which they were trained. My work combines reinforcement learning, language-model reasoning, and formal structure to build agents that are more capable, transferable, and reliable.

About me

I am a fourth-year Ph.D. student and Graduate Research Assistant in the School of Computer Science at the Georgia Institute of Technology, advised by Dr. Suguman Bansal. My research focuses on reinforcement learning, LLM-based agents, and generalization, particularly on agents that must solve families of related tasks rather than a single fixed problem.

I develop learning methods that exploit task structure and formal specifications to help policies generalize to new tasks and environments. A central theme of my work is inductive generalization. This involves learning reusable policies from a limited set of training tasks while systematically transferring them to larger or unseen task families.

I am also interested in reinforcement learning for LLMs and agentic systems, including post-training, reasoning, planning, and decision-making in interactive environments. More broadly, I am interested in combining LLM reasoning, reinforcement learning, and formal methods to build agents that can adapt to new problems while remaining reliable and verifiable.

Research interests

  • Reinforcement learning generalizationScalable policies that transfer across related tasks and environments.
  • LLM agents and planningLanguage-model agents for reasoning, decision-making, and interactive tasks.
  • Learning from specificationsUsing task structure and logical descriptions to guide policy learning.
  • Evaluation and certificationUnderstanding when learned behavior generalizes and where it fails.

News

  • May 2026Our papers, Decoupled Behavioral Cloning for Scalable Inductive Generalization in RL from Specifications and Certificate-Guided Evaluation of Reinforcement Learning Generalization, are now available as arXiv preprints.
  • Apr 2026Generalization in Reinforcement Learning from Logical Specifications was accepted to the KR 2026 Doctoral Consortium.
  • Apr 2026Gave a research talk on Decoupled Behavioral Cloning at the PLSEFM + AI Workshop at Georgia Tech.
  • Jan 2026Gave a research talk on Inductive Generalization in Reinforcement Learning from Specifications at IIT Kanpur.
  • Jul 2025Inductive Generalization in Reinforcement Learning from Specifications was accepted at ATVA 2025.
  • Jun 2025Certificate-Guided Evaluation of Reinforcement Learning Generalization was accepted at SAIV 2025 as a presentation-only paper.
  • Dec 2024Certification-Guided Evaluation of Reinforcement Learning Generalization was accepted for the AAAI GenPlan 2025 Workshop.
  • Jun 2024Inductive Generalization in Reinforcement Learning from Specifications became available as an arXiv preprint.
  • Apr 2024Won third place in the Graduate Poster Symposium, Junior Student Category.
  • Oct 2023Inductive Generalization in Reinforcement Learning from Specifications was accepted at the NeurIPS GenPlan 2023 Workshop.
  • Aug 2023Joined Georgia Tech as a Ph.D. student under Dr. Suguman Bansal.
  • May 2023Reinforcement Learning for Stochastic Max-Plus Linear Systems was accepted at CDC 2023.
  • Mar 2023A novel facial emotion recognition model using segmentation VGG-19 architecture was published in IJIT.
  • May 2022SIHeDA-Net was accepted at MIDL 2022.
  • Jan 2022Started a research internship at Newcastle University working on Deep Q-Learning for stochastic max-plus-linear systems under uncertainty.
  • Aug 2021Started a research internship at Samsung R&D Institute, Bangalore working on occlusion handling for normal-map estimation from degraded 2D images.
  • Feb 2021Started a research internship at NIT Tiruchirappalli working on a U-Net and VGG-based model for facial emotion recognition.