purpose.

S.M. Obaydur Rahman is a full-stack software developer with over four years of professional experience architecting production-grade systems at the intersection of banking, fintech, and applied machine learning. He is currently employed at Mutual Trust Bank PLC (MTB) within the Digital Banking Division — Satellite & Ecosystem Unit — where he leads the design and implementation of core banking middleware, open-banking REST APIs, and AI-powered automation workflows. His long-term goal is to pursue a fully-funded PhD that fuses the empirical discipline of production engineering with rigorous research in deep learning, natural language processing, and trustworthy AI for high-stakes, real-world domains.

Why software engineering — and who it serves. Bangladesh is a country of roughly 170 million people, the majority of whom interact with the formal financial system through a thin, fragile layer of middleware: mobile financial services (MFS), agent banking, remittance corridors, and bill-collection rails. Every API he writes at MTB sits squarely on that layer. The payment gateway integrations he maintains move salaries to garment workers, remittances from migrant workers in the Gulf and South-East Asia back to rural households, and utility payments for families who have never set foot inside a branch. The dispute-automation platform he built on .NET 9 shortens the resolution window for customers who would otherwise wait weeks for a card-transaction reversal. The Knowledge Portal he engineered gives frontline operations staff — many of them young women working their first formal job — instant, accurate answers in place of buried PDFs and tribal knowledge. In other words, the software is not abstract: it determines whether financial services in a Global South economy are inclusive, auditable, and fast, or exclusionary, opaque, and slow.

From production engineering to research. Building these systems has shaped his research mindset in three concrete ways. First, working under regulatory scrutiny — Bangladesh Bank guidelines, PCI-DSS, AML/CFT, KYC — has made him uncompromising about data integrity, reproducibility, and auditability, the same qualities that distinguish credible ML research from brittle demos. Second, debugging real customer incidents at 2 AM has taught him that distributional shift, edge cases, and silent failures are not abstractions in a textbook but the actual frontier where models meet the world. Third, integrating LLM-based assistants and OCR pipelines into a bank's operational workflow has convinced him that the most consequential AI research over the next decade will not be benchmark-chasing, but rather trustworthy, interpretable, and resource-efficient systems that hold up under audit, latency budgets, and adversarial conditions. This is the worldview he wishes to bring into a doctoral program.

Why deep learning, ML, and AI. His technical portfolio spans the full stack — Oracle and SQL Server schemas, Java Spring Boot, .NET 9 / C#, PHP / CodeIgniter / Laravel, Node.js, React, and Flutter — but his intellectual centre of gravity has always been data-driven inference. As an undergraduate at Shahjalal University of Science and Technology (SUST), his thesis applied classical machine learning, ensemble classifiers, and LSTM with Word2Vec embeddings to sentiment analysis of COVID-19 post-vaccination discourse in Bengali, achieving 78.8% best-case accuracy. That project surfaced a problem he has not been able to put down: Bengali, the seventh-most-spoken language on earth, is still treated as a low-resource language in mainstream NLP. Pre-trained tokenizers fragment its morphologically rich script, public datasets are scarce, and benchmark leaderboards are dominated by English and Mandarin. The same asymmetry holds for Bangla healthcare records, agricultural advisories, and legal documents — domains where the social return on better models is enormous and the academic neglect is acute.

That motivation matured into his most recent contribution, presented at the IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence and Networking (QPAIN 2026): "Advanced Hybrid Deep Learning Models for Solar Energy Holding Capacity Prediction in Bangladesh." The study introduces an ensemble framework integrating Random Forest, XGBoost, Gradient Boosting, AdaBoost, and a Neural Network meta-learner, achieving roughly 95% predictive accuracy on Bangladeshi solar irradiance data. The work is not just a methodological exercise — it speaks directly to the energy-planning crisis of a country that imports most of its primary energy and is acutely vulnerable to climate shocks. He sees AI for climate and renewable energy in the Global South as one of the defining research agendas of his generation, and one where local researchers with local data have a structural advantage.

Current research interests. His doctoral interests cluster around four tightly-coupled themes: (i) hybrid and ensemble deep learning — combining classical statistical learners with neural meta-models for tabular, time-series, and small-sample regimes typical of emerging-economy datasets; (ii) NLP for low-resource languages, with a specific focus on Bengali morphology, code-mixed Bangla-English text, and domain-adapted LLMs for South Asian contexts; (iii) AI for climate, energy, and sustainability, particularly forecasting, optimization, and decision-support for renewable-energy planners and policymakers; and (iv) trustworthy ML for high-stakes domains — fraud detection, AML/KYC, credit scoring, and clinical decision support — where interpretability, fairness, and robustness are not optional. He is equally interested in the systems side: retrieval-augmented generation (RAG), efficient fine-tuning (LoRA / QLoRA), agentic workflows, and on-device inference for bandwidth-constrained environments.

Credentials and recognition. Rahman holds a B.Sc. in Computer Science and Engineering from SUST, Sylhet (CGPA 3.06 / 4.00). He is an Associate Member of both the Institution of Engineers, Bangladesh (IEB, A-28590) and the Bangladesh Computer Society (BCS, AM16230), and is entitled to use the professional designation "Engr." He was recognised in the Top 100 Teams at the SOLVIO AI Hackathon 2025 for an agentic payment-automation solution, and has completed graduate-level coursework in Supervised Machine Learning, Applied Machine Learning, Exploratory Data Analysis, and ITIL v4 service management. He maintains active open-source contributions and a growing portfolio of fintech and AI side-projects.

Looking forward. Rahman is actively seeking a fully-funded PhD position with a supervisor whose research agenda intersects with hybrid deep learning, low-resource NLP, AI for sustainability, or trustworthy ML in finance or healthcare. He brings to a doctoral program something unusual: not just coursework and a publication, but four years of evidence that he can ship, maintain, and reason about complex systems that real people depend on every day. He welcomes conversations with prospective supervisors, collaborators, and industry partners who are working on problems where careful, long-form research engagement matters.