Zaidur Rahman
Data Scientist and Data Engineer
I build data pipelines, machine learning models and decision tools. I'm a PhD candidate in Industrial Engineering at the University of Arkansas, and before that I worked in supply chain and project management
I'm looking for Data Scientist, Supply Chain and Industrial Engineer roles.
About me
I grew up in Bangladesh and studied Industrial and Production Engineering at Khulna University of Engineering and Technology. Outside class I spent a lot of time building robots, including line follower robots and one that solves a Rubik's Cube. That habit of taking a problem apart, finding the part that can be measured, and building something that acts on it has stayed with me.
My first job was in supply chain at Gildan, and it changed how I think about data and models. The hard part was getting reliable data out of people and systems that didn't talk to each other, and then giving the team something they would actually trust and use.
Then I came to the U.S. for graduate school, first at North Dakota State and now at the University of Arkansas. My work has moved steadily from operations research toward machine learning, data engineering and AI agents. I'm most interested in the space between research and practice: taking a method that works on paper and making it work on real, messy data.
Outside study and research, I travel, camp, hike, and dive, and I always come home to my wife and our fur babies, Leo and Smokey.
How I work with data
Collect
Large-scale geospatial vessel-tracking data and federal waterway shapefiles.
Store and clean
Pipelines into a PostGIS database, and a graph model of the river network.
Model
ETA prediction using AI agents, text-to-SQL agent to interact with massive data, and stochastic optimization.
Decide
Data retrieval, ETA windows and service schedules, benchmarked against first-come-first-served.
University of Arkansas, 2024 to now
Collect
Order, delivery and capacity data from vendors, the warehouse, transport and the production floor.
Store and clean
One real-time, web-based scheduling system that unified the information flow between all four.
Model
Reorder planning, aggregate planning and a priority-based warehouse system.
Decide
15% more daily yarn on the floor, 20% fewer stockouts, 35% less material handling time.
Gildan Activewear, Bangladesh, 2019 to 2021
Collect
Labor hours, equipment profiles, bills of lading (BOL), task progress from each department, proofs of delivery and equipment records.
Store and clean
Automated data pulls into the BOL matrix and automated Plan of the Day (POD) data entry.
Model
A lost-hour calculation app that breaks hours down by department and cost code.
Decide
Equipment tracking that cut downtime by an average of 9.7 hours per week.
Moss & Associates, solar project, Arcadia, Florida, 2025
Collect
Python scrapers pulling protein data from the Database of Essential Genes, BLAST and PSORTb.
Store and clean
An automatically built, labeled training dataset with engineered features.
Model
Random Forest, XGBoost and AdaBoost classifiers, plus network analysis to cluster functions.
Decide
Predicted functions for 83 unannotated proteins, 8% more accurate than the PANNZER tool.
North Dakota State University, 2021 to 2023
