Graphs change every second. I build parallel algorithms and GPU data structures that update only what changed.
High performance computing, dynamic graphs and hypergraphs, and scalable AI. PhD from Missouri S&T, with research stints at PNNL, Fermilab, and LLNL.
Real networks never sit still. Roads congest, people connect, and data keeps streaming in. Instead of recomputing everything after each change, I find exactly what a change touched and repair only that, in parallel, on multicore CPUs and GPUs. I carry this idea from hypergraph data structures to wildfire evacuation routes and scalable machine learning.
Postdoctoral Scholar, AI(X) Hub, The Ohio State University, with Prof. Ness B. Shroff
Computer Science, Missouri S&T, 2026. Advised by Prof. Sajal K. Das
PNNL 2024
Fermilab 2025
LLNL 2026
NSF I-Corps lead, editorial board member, open-source CANDY contributor
Joined the AI(X) Hub at The Ohio State University as a Postdoctoral Scholar.
Defended my PhD at Missouri S&T and completed an internship at Lawrence Livermore National Laboratory.
ESCHER+ accepted in IEEE TKDE; shortest hyperpath update accepted at the IA³ Workshop, SC26.
Presented ESCHER at IEEE IPDPS 2026 in New Orleans. DynLP appeared at ACM ICS 2026 and RESCUE at ICDCN 2026.
Citations and h-index from Google Scholar, Sept 2026
Each line of work starts from the same question: when the input moves, what is the least work needed to keep the answer right?
ESCHER is the first GPU data structure for dynamic hypergraphs. It absorbs batches of hyperedge insertions and deletions and keeps triad counts current. ESCHER+ makes it lean on memory.
Parallel algorithms that update multi-objective shortest paths and shortest hyperpaths after each batch of changes, and RESCUE, which routes wildfire evacuations under evolving congestion and uncertain spread.
DynLP updates label propagation for graph-based semi-supervised learning in parallel batches, instead of re-running it after every change. It grew out of my internship at PNNL.
At LLNL I sped up SOLANET's approximate nearest neighbor graph construction on AMD MI300A and NVIDIA H100. At Fermilab I studied ROOT RNTuple storage layouts for the DUNE experiment across 20 data product configurations.
Before HPC, I worked on machine learning for biology and health: protein post-translational modification sites, cancer biomarkers, and Alz-Sense+, a dementia detection algorithm that became the basis of an NSF I-Corps team. That work accounts for 17 of my papers.
Postdoctoral Scholar, AI(X) Hub · Columbus, OH
Summer Intern · Livermore, CA · Mentors Keita Iwabuchi and Min Priest
FCSI Intern · Batavia, IL · Mentor Philippe Canal
PhD Summer Intern · Richland, WA · Mentors S M Ferdous and Mahantesh Halappanavar
Graduate Research and Teaching Assistant · Rolla, MO · Advisor Prof. Sajal K. Das
Lecturer, Computer Science & Engineering · Rajshahi
Lecturer, Computer Science & Engineering · Dhaka
PhD, Computer Science
Missouri S&T · GPA 4.0/4.0
MSc, CSE
RUET · CGPA 3.75/4.00
BSc, CSE
RUET · CGPA 3.96/4.00, first merit position
Presidential Gold Medal
Conferred by the President of Bangladesh for the highest CGPA across all departments, BSc
Best Student Award
RUET, awarded by the Vice-Chancellor for academic excellence
Editorial board
Cloud Computing and Data Science
Peer review
30+ manuscripts for SC, ACM HPDC, Scientific Reports, Cluster Computing, BioData Mining, Molecular Diversity, Discover AI, Peer-to-Peer Networking & Applications, two Frontiers journals, and JKSU Computer and Information Sciences.
Parallel
Languages
Profiling & ML