Before that, I was a Graduate Research Assistant in the Distributed Intelligent Systems (DISys) Laboratory, advised by Dr. Paritosh Ramanan, where I worked on decentralized optimization for decarbonizing energy-intensive industry. I continue that work by building an agentic AI framework that tunes the penalty parameter of decentralized optimization algorithms, so they converge efficiently without hand tuning.
Earlier, I spent three and a half years as a data analyst in industry and taught industrial engineering as a lecturer. I am now applying to PhD programs in Industrial Engineering and Operations Research for Fall 2027.
Research
The curiosity behind my research is a question I keep returning to: can independent systems, each guarding its own data, still decide together as well as if they had shared everything? Two directions of my work pursue it directly. Alongside them, I work on healthcare analytics, where the challenge shifts from coordination to extracting reliable signals from large, messy clinical data.
01
Decentralized decision-making under uncertainty
Power grids and the industrial plants they serve must plan together, but plants will not share their operating data. I build methods that let them coordinate anyway.
Coordinate
We developed a decentralized framework in which a 2,000-bus transmission grid and 26 electrified chemical plants schedule jointly through ADMM, exchanging only coordination signals instead of private operating data.
Handle uncertainty
My thesis adds uncertain on-site wind and solar. Each plant becomes a two-stage stochastic program, the grid and plants negotiate reserve capacity rather than power demand, and renewable scenarios are generated without pooling plant data.
Forecast
Federated learning carries the same principle into prediction. Plants train a shared load forecaster while their meter data stays local, so a newly connected plant with little history still receives accurate forecasts.
Decentralized algorithms such as ADMM converge only as fast as their penalty parameter allows, and that parameter is usually tuned by hand for each problem.
I am developing skills for Scarlet Composer, our lab's multi-agent framework, that run the grid and each plant as independent agents solving their own subproblems. An LLM coordinator observes only privacy-preserving convergence signals and adapts the penalty, so coordination tunes itself across problems without any plant revealing its data.
Real-world clinical records covering more than 10 million patients, used to map how comorbidities emerge before and after a Juvenile Idiopathic Arthritis diagnosis to support earlier detection, alongside machine-learning models for disease prediction.
Real-world dataTrajectory modelingPredictive ML
Also working on
Agentic AI evaluation
Frameworks that test conversational analytics agents against automatically generated reference questions, catching hallucinated answers before deployment.
M.S. Business Analytics and Data Science, GPA 4.0 Oklahoma State University. Graduate Certificate in Business Analytics and Data Science awarded 2025.
2024 to 2026
M.S. Industrial Engineering and Management, GPA 4.0 Oklahoma State University. Operations Research and Analytics concentration, thesis option. Advisor: Dr. Paritosh Ramanan.
2016 to 2020
B.S. Industrial and Production Engineering, GPA 3.58 Military Institute of Science and Technology, Dhaka
Experience
since 2026
Advanced Analytics Intern, Data Research and Analytics, AAA, The Auto Club Group Conversational data agents with Google ADK and Gemini over BigQuery and Tableau, Dataplex data-quality checks, and a marketing-attribution model with Google Meridian.
2025 to 2026
Graduate Research Assistant, Distributed Intelligent Systems (DISys) Laboratory, Oklahoma State University Privacy-preserving stochastic reserve coordination between a 2,000-bus grid and 26 chemical plants via two-phase ADMM on HPC clusters (cost within 1.3% and emissions within 0.7% of a centralized solution), and trajectory-modeling pipelines on real-world clinical data.
2024
Graduate Teaching Assistant, Oklahoma State University, IEM Python and Excel labs for ENGR 1412, Canvas for 60+ students, roughly 95% positive evaluations.
2021 to 2024
Data Analyst, LandQuire Management LLC Automated ETL and dashboards with Python, SQL, and Supermetrics, cutting manual reporting by 85%. Certificate of Excellence recipient.
2020
Lecturer, European University, Bangladesh Operations research, manufacturing, and optimization; mentored 30+ capstone students.
Honors and awards
Awards and competitions
2026
National Champion, Data4Good Competition, Johns Hopkins University Press release
2026
Finalist (Top 5), IISE Energy Analytics Challenge
2026
1st Runner-Up, OSU Datathon, Spears School of Business
2026
Winner, M.S. BAnDS Business Case Competition, Oklahoma State University
Research Symposium Token Prize, 2nd Annual IEM Research Symposium, Oklahoma State University
2024
Certificate of Excellence, LandQuire Management LLC
Scholarships
2026
Creativity, Innovation and Entrepreneurship (CIE) Scholar, Oklahoma State University
2025
SAS Advisory Board Scholarship, Oklahoma State University
2017, 2018
Merit Scholarship, Military Institute of Science and Technology Top 5 of the academic batch
Honor societies
2026
Alpha Pi Mu, Industrial Engineering Honor Society
2025
Tau Beta Pi, Engineering Honor Society
Leadership and service
since 2025
GPSGA Representative, Business Analytics Student Association, Oklahoma State University
2024 to 2025
Vice President, INFORMS Student Chapter, Oklahoma State University
ongoing
Reviewer, Expert Systems with Applications and Healthcare Analytics (Elsevier)
Selected projects
Blockchain
GridFlex ChainBesu/QBFT, IPFS. A permissioned blockchain market where homes, critical facilities, and an industrial microgrid trade surplus solar through sealed-bid auctions, with commit-reveal privacy for meter readings and bids, priority allocation for critical loads, and a tamper-evident audit trail. In progress.
Optimization
Power Flow Optimization Web AppPython, Pyomo, Gurobi, Streamlit. Reduces generation cost by 15% and visualizes grid operations with auto-layout diagrams. Live · Code
Optimization
Optimized Work SchedulingMILP, Gurobi, Python. Cut scheduling conflict hours by 90% with 100% shift coverage. Code
NLP
Ask-YouTube ChatbotEmbeddings, retrieval, Streamlit. Timestamp-aware question answering over video transcripts. Live · Code
Acknowledgments
I am grateful to the mentors and organizations that have supported my research and growth.
Mentors: Dr. Paritosh Ramanan (Research Advisor), Dr. Jennifer Glenn, Dr. Akash Deep, Dr. Zheyu Jiang, Dr. Mohammad Fili, Dr. Guiping Hu, Dr. Lizhi Wang, and Dr. Dursun Delen, and my undergraduate mentors at the Military Institute of Science and Technology.
My research is supported in part by the National Science Foundation (NSF) under Award 2343072. Another part of my research is supported under the NASA HOME STRI Project (Research Thrust 2) and NSF SaTC Award 2348411.
Contact
The best way to reach me is by email. I am glad to talk about optimization under uncertainty, agentic analytics, and research collaborations.