Summary
Amit Roy is a PhD student at Purdue CS and an active researcher specializing in transformer-based LLMs, LLM agents, and graph-based deep learning with eight years of experience spanning academia and industry. He investigates multi-hop reasoning, compositional generalization, and interpretability in vector spaces to enable LLMs to search, retrieve, update, and forget information, with work published at venues including WSDM and NeurIPS workshops. His applied research includes dynamic graph link prediction, graph anomaly detection, and spatio-temporal forecasting, and he has built retrieval and code-executing agent systems during an Applied Scientist internship at Amazon. Amit combines theoretical depth with practical ML engineering from prior roles at Futurewei, TigerIT, and Independent University Bangladesh, and he was awarded the Ross Fellowship at Purdue. Notably, he blends LLM techniques with GNNs—e.g., distilling LLM semantic representations into efficient GNNs for dynamic text-attributed graphs—aiming to instill human-like reasoning into agents.
8 years of coding experience
3 years of employment as a software developer
Secondary School Certificate (SSC), Science, Secondary School Certificate (SSC), Science at Syed Sayeed Uddin High School
Master of Science , Computer Science, Master of Science , Computer Science at Purdue University
Bachelor of Science - BSc (Honours), Computer Science and Engineering, Bachelor of Science - BSc (Honours), Computer Science and Engineering at University of Dhaka
Higher Secondary Certificate, Science, Higher Secondary Certificate, Science at Rajuk Uttara Model College
Hindi, English, Bengali