👋 Hello there, I’m Nazrul!

🎓 I’m a Graduate Research Assistant under Dr. Paritosh Ramanan at Oklahoma State University. I am working on a NASA HOME STRI associated project to develop an AI-agent orchestration framework using LangChain, MCP, and Gustavo for decentralized IoT systems with privacy-preserving verification. Previously, I worked under Dr. Sharmin Jahan, where I built a dataset from microservices traffic and developed machine learning models for anomaly detection, creating a self-protective service mesh that automatically handles attacks.

💼 With over 11 months of experience as a Junior Software Engineer at Tirzok Private Ltd., I am specialized in backend development. My work ranged from developing AI solutions and integrating AWS services to building backend systems using clients’ preferred frameworks and databases.

📚 I also have over two years of experience as an Undergraduate Research Assistant at BRAC University, where I honed my skills in data collection and analysis.

🔍Currently, I am actively seeking jobs starting from Summer 2026. My interest in research and technology grew under the mentorship of Dr. Jannatun Noor, who taught me to be flexible and adapt to emerging technologies every day. I am excited to bring my background in software engineering, infused with AI expertise and research experience to innovative teams where I can contribute, learn, and grow.

Feel free to explore more of my work and projects at my portfolio and CV. Please don’t hesitate to connect if you see potential for collaboration or simply wish to network.

If you are viewing the homepage from a mobile phone, please click the menue on the top-right corner to find other navigations. Furthermore, you can find contacting information by clicking follow button.

Selected Experience

Automatic Document Validation with Textract

For a US-based borrowing company Kramasoft, I implemented automatic validation for borrower documents by developing machine learning classifiers to identify document types, enhancing the system’s verification process. I engineered a specialized document trimmer that uses advanced ML classifiers to intelligently remove irrelevant pages, sending only important pages to AWS Textract for precise data extraction. I developed robust extraction processes to retrieve crucial field values accurately and built RESTful APIs using Spring Boot and PostgreSQL to facilitate auto-validation processes. Additionally, I integrated AmazonMQ (RabbitMQ) for seamless internal communication and deployed the system in AWS Lambda using ECR.

I built an AI-driven self-protective service mesh that combines anomaly detection with automated defense for microservices. To support this, I collected and engineered a dataset of over 1.2M Istio/Envoy traffic logs and trained ML models with Scikit-Learn, achieving 92% DDoS detection accuracy with LIME explainability. When attacks were detected, the system automatically removed affected services and rerouted traffic to healthy alternatives, creating a self-healing, resilient cloud-native environment powered by Kubernetes, Istio, and Docker.

LangChain + MCP Chatbot Framework

I built a chatbot framework with a Bootstrap frontend and Flask backend that connects to LangChain’s ReAct agent for multi-service orchestration. The system integrates MCP tool servers (e.g., a weather API and math operations) and leverages GroqIllustration of combining vision and language modalities for fast LLM inference, allowing the chatbot to dynamically select and combine services in real time. This project demonstrates how LLMs can act as intelligent orchestrators, coordinating multiple APIs and tools within a seamless conversational interface.

Bridging the Digital Divide: A Study of HVECs in Southeast Bandarban

In the high mountains of southeast Bandarban, highly vulnerable ethnic communities (HVECs) live without basic mobile and network availability. To address the digital divide among HVECs, we conducted a mixed-method study involving two visits Illustration of combining vision and language modalities to six different ethnic communities across 15 villages. Trekking for nine days to collect data, we engaged 72 participants in quantitative and qualitative research, using statistical and machine learning methods. Our findings revealed that 34.2% of participants had never used the Internet, and only 13.7% used it once a week. We explored their Internet skills, outcomes, and reasons for limited use, providing insights to narrow the physical access divide and inform future HCI4D design. This study contributes to understanding the digital divide and technology design for distant indigenous communities, broadening the scope of the HCI4D/ICT4D community.

Developing the Backend of Janatar Sarkar

I developed RESTful APIs for the backend of the Janatar Sarkar government-public interaction website using Node.js and MongoDB. I also implemented a comprehensive Role Management System utilizing JWT authentication, Node.js, and MongoDB. This system ensures secure access control tailored to various types of administrators within janatarsarkar, allowing precise assignment of roles and responsibilities