My research focuses on accelerating the discovery of thermoset shape memory polymers (TSMPs) using generative artificial intelligence. I develop deep learning–based generative models, large language models (LLMs), and multi-stage reward systems to create polymer structures that align with target properties such as Tg and elastic recovery. This work aims to replace slow, trial-and-error experimentation with a scalable AI-driven design loop.
I am currently developing an agentic AI framework for polymer and materials discovery that integrates several coordinated components:
Generative Models: Deep generative architectures and LLMs that propose new two-monomer TSMP candidates based on desired thermal and mechanical properties.
Reward and Feedback Models: A combination of rule-based scoring, human feedback, AI judgment models, and chemistry-aware validity checks to guide the generator toward realistic and novel materials.
Value and Critic Agents: Models that evaluate manufacturability, functional group compatibility, and underlying chemical mechanisms.
Synthesis & Feasibility Agents: RDKit- and SMARTS-driven analysis, along with external planning tools, to assess chemical synthesizability and reaction pathways.
Evaluation Agents: Automated systems that compare generated polymers with ground truth, measure diversity, ensure novelty, and validate chemical plausibility.
Together, these components form the foundation of a closed-loop agentic AI system—one that can autonomously propose TSMPs, refine them through feedback, and continuously improve over time. My long-term goal is to build a flexible agentic AI framework capable not only of advancing materials discovery but also of extending to healthcare and other scientific domains where generative reasoning and automated evaluation are essential.
Alongside materials research, I have worked in healthcare AI, including computer vision methods such as automated mask detection during the COVID-19 pandemic. I plan to expand this direction further by applying generative AI, large language models, and vision systems to healthcare tasks—for example, generating synthetic medical data for training, analyzing medical images for anomaly detection, and building LLM-based triage or diagnostic-assistance tools.
Looking ahead, my goal is to connect these two domains through a unified agentic AI framework. The same generative and evaluative agents used for polymer discovery can be adapted to explore bio-materials, drug-delivery polymers, wound-healing materials, or diagnostic coatings. Similarly, LLM-based reasoning and vision-based evaluation modules can support both materials characterization and healthcare decision systems. By extending my agentic AI architecture across these areas, I aim to build adaptable tools that advance innovation in both materials science and health-oriented AI applications.
Current Research
1. Reinforcement Learning–Based Reward Model for Polymer Generation
I am developing a reinforcement learning–driven reward model to improve the quality and realism of thermoset shape memory polymers (TSMPs) generated. This system integrates rule-based chemical scoring, human and AI feedback signals, property-based evaluations, and novelty/diversity metrics. The reward model guides a generative agent to produce polymer structures that better match target properties such as Tg and elastic recovery, while maintaining chemical validity and synthesizability. This work forms a core component of a self-improving generative loop for materials discovery.
2. Agentic AI Framework for Materials Discovery
I am building a multi-agent AI framework capable of autonomously generating, assessing, and refining candidate polymers. The system coordinates several specialized agents—generative agents, critic agents, synthesis-feasibility agents, and evaluation agents—to form a closed-loop discovery pipeline. Each agent performs a specific scientific reasoning or validation task, enabling the framework to propose new TSMP candidates, critique them, and iteratively improve performance. This agentic AI system is the foundation for a broader, extensible platform that can be adapted to other domains, including healthcare-related material design and diagnostic applications.
Grants & Proposal Writing
Assisted in the preparation of two research proposals that are currently under review.
Contributing to two additional proposals under development with Prof. Hei, University of Louisiana at Lafayette.
Support includes background research, AI methodology design, and documentation.
Publications
2025
Peer-Reviewed Journal Articles
1. Borun Das, Andrew Peters, Guoqiang Li, and Xiali Hei, "Prompt2Poly: Ask, Specify, Create – A Dialogue-Based Large Language Model for Targeted Polymer Design.", Polymer Chemistry, Royal Society of Chemistry (Impact Factor: 3.9), 2025.
2. Borun Das, Andrew Peters, Guoqiang Li, and Xiali Hei, "Generative Design of Thermoset Shape Memory Polymers Driven by Chemical Group: A Conditional Variational Autoencoder Approach.", Journal of Polymer Science, John Wiley & Sons (Impact Factor: 3.6), 2025.
2024
Conference Proceedings & Technical Papers
1. Borun Das, Guoqiang Li, and Xiali Hei, "Chemical Group-Driven Generation of Multi-Monomer Thermoset Shape Memory Polymers Using Generative Conditional Variational Autoencoder." , in Proceedings of the Louisiana Materials Design Alliance (LAMDA) Symposium, 2024.
2023
Conference Proceedings & Technical Papers
1. Borun Das, Guoqiang Li, and Xiali Hei, "Unveiling Thermoset Shape Memory Polymers through RNN. , in Proceedings of the LAMDA Symposium, 2023.
2022
Conference Proceedings & Technical Papers
1. Borun Das, Mahmud Hasan, and Rafi Belal, "A Deep Learning Approach for Detecting Medical Face Masks on Human Faces in Response to COVID-19.", in Proceedings of the 11th International Conference on Robotics, Vision, Signal Processing and Power Applications, Springer, 2022.