Subhajit Paul

Research

What I work on, what I've published, and what's in progress.

Research Interests

I work on making hardware verification more reliable and more automatic. Two questions drive most of it: can LLM agents write and repair the assertions that verification engineers currently write by hand — and how do multiple learning agents coordinate when they share an environment? The first is my day-to-day work at ISI Kolkata; the second is multi-agent reinforcement learning, the direction I want to take a PhD in.

News

  1. 2026

    SuperSAGA accepted at ASP-DAC 2026 (Special Sessions).

  2. 2026

    VISTA submitted to the ACM Journal on Emerging Technologies in Computing Systems.

  3. 2025

    LLM-Guided Reconciliation for Explainable Robotic Path Planning accepted at IEEE ACSOS 2025.

  4. 2025

    LISA accepted at IEEE ISVLSI 2025.

  5. Jul 2024

    Joined the Indian Statistical Institute, Kolkata as a Research Engineer.

Publications

Formal Verification and Agentic AI

Subhajit Paul, Ansuman Banerjee, Sumana Ghosh

Accepted at IEEE Computer Society Annual Symposium on VLSI (ISVLSI 2025)

Indian Statistical Institute, Kolkata

Code

Subhajit Paul, Ansuman Banerjee, Sumana Ghosh

Accepted at 31st Asia and South Pacific Design Automation Conference (ASP-DAC Special Sessions 2026)

Indian Statistical Institute, Kolkata

VISTA: An Agentic Workflow for Formal Verification Environment Synthesis

2026

Subhajit Paul, Ansuman Banerjee, Sumana Ghosh

Under Review and Submitted to the ACM Journal on Emerging Technologies in Computing Systems

Indian Statistical Institute, Kolkata

LLMs in Robotics

Subhajit Paul, Ansuman Banerjee

Accepted at IEEE Autonomic Computing and Self-Organizing Systems (ACSOS 2025)

Indian Statistical Institute, Kolkata

Ongoing Research Work

Multi-Agent Reinforcement Learning

Exploring cooperative and competitive learning among autonomous agents, focusing on coordination, credit assignment, and emergent behaviour in complex multi-agent settings.

LLM-based Agentic Systems

Building tool-augmented, supervisor-subordinate LLM agents that plan, reason, and self-correct to automate formal verification and environment synthesis within EDA pipelines.

Resources I Learn From

Buy Me A Coffee Icon