Sourin Dey
PhD candidate in Computer Science at the University of South Carolina, building AI for materials science and AI systems that companies can run in production.
Agentic AI & applied systems
I build AI systems that organizations can actually run. I understand what it takes to move an AI workflow from prototype to production: orchestrating models and tools, data pipelines that keep knowledge current, deployment, and guardrails that keep outputs trustworthy. In one industry role, I built and deployed an AI product selector for sales teams that draws on decks, video and SharePoint data through an automated ingestion pipeline. In another, I built a graph-RAG assistant over decades of software manuals with LangChain and Neo4j. I work with LangGraph, Pydantic AI and LangChain.
Research
My PhD research uses AI to speed up materials discovery, with a focus on building the underlying physics into neural-network models. Current threads: fast crystal structure generation with mean flow models, mapping crystal structures through polyhedral topology, accurate band gap prediction with Δ-learning, and active learning for laser-induced graphene manufacturing. I have six years of academic and industrial experience with predictive and generative models and active-learning pipelines.
