Miraj Uddin Chowdhury

NLP Researcher and Software Developer

Computer Science graduate and NLP researcher focused on resource efficient language models, model compression, and low resource Bangla processing. I work with knowledge distillation, LoRA, quantization, RAG, OCR, and enterprise NLP workflows to build practical AI systems with efficient inference and reliable software engineering.

Explore Projects Get In Touch
Miraj Uddin Chowdhury

About Me

Computer Science graduate and NLP researcher specializing in resource efficient Large Language Models and model compression. My research experience includes knowledge distillation, parameter efficient fine tuning with LoRA, model quantization, and compact architectures for Bangla and other low resource language settings. I am interested in low latency language modeling, efficient representation learning, and dependable AI systems.

🎓 I completed my Bachelor of Science in Computer Science and Engineering at International Islamic University Chittagong.

🔬 My current research focuses on compact language models for Bangla cyberbullying and threat detection using knowledge distillation, LoRA, hidden layer matching, and dynamic quantization.

💻 As an Associate Software Developer at Akij iBOS, I contribute to LLM and NLP workflows for business process automation, while maintaining tested code and clear technical documentation.

📚 My research interests include Natural Language Processing, efficient language models, low resource language processing, RAG, vector retrieval, model compression, and efficient representation learning.

Halishahar, Chattogram, Bangladesh
+8801830484860
mirazchowdhury03@gmail.com
BSc in Computer Science and Engineering
NLP, Tiny LLMs, Knowledge Distillation, LoRA, Quantization

Technical Skills

Programming and Tools

Python SQL Git GitHub LaTeX MS Excel Google Sheets

AI and NLP

Large Language Models Natural Language Processing Knowledge Distillation PEFT LoRA Quantization RAG Vector Search

Machine Learning

TensorFlow Scikit Learn Text Classification TF IDF Model Evaluation Prompt Engineering

Data Analysis

Pandas NumPy Matplotlib Seaborn Data Cleaning Data Visualization Automation Pipelines

Document Intelligence

Docling Tesseract OCR OpenCV Regex Structured JSON PDF Extraction

Research and Quality

Academic Research Technical Documentation Problem Solving Software Testing JMeter Selenium Postman

Experience

Akij iBOS

Associate Software Developer

January 2026 to Present

Ongoing
  • AI and NLP Development: Implemented Large Language Model and Natural Language Processing workflows for enterprise business process automation.
  • Software Engineering: Contributed to AI projects under senior engineers, with attention to maintainable code, testing, and documentation.
  • Applied Research: Studied emerging AI methods and translated relevant ideas into practical technical solutions for operational efficiency.

International Islamic University Chittagong

Undergraduate Research Assistant, Department of CSE

Supervised by Prof. Md. Mahiuddin

Research Role
  • Conducted research on embedded IoT systems and low latency transaction reporting for a campus payment infrastructure.
  • Designed Raspberry Pi and RFID system integration, and documented structural data flows and analytical results.

NexGenix Department of AI Research and Development

Data Analyst Intern

September 2025 to November 2025

Completed
  • Cleaned and processed large datasets using Python for AI research and analytics workflows.
  • Automated preprocessing and reporting tasks, reducing manual effort and limiting avoidable processing errors.

Featured Projects

PixelRAG PDF Chatbot

Python, FastAPI, Ollama, Docling, ChromaDB, SQLite, RAG

2026, Local Multilingual Document Assistant

Developed a local multilingual PDF assistant that supports multiple document uploads, document based question answering, page citations, and supporting evidence images. Built a modular processing workflow with Docling, local Ollama models, persistent ChromaDB vectors, SQLite metadata, voice and text input, and Bengali, English, and Banglish support. The current version includes a testing retriever for GPU independent workflow validation, while the PixelRAG adapter remains ready for visual embeddings, Qdrant, and multimodal model integration.

View Project

NeuroScope AI

Python, Groq Cloud API, Gradio, LLMs, Agentic Workflows

Agentic AI Chatbot Application

Architected a real time agentic chatbot with autonomous web search tool calling and custom domain filtering. Added conversational state management, tailored prompting, asynchronous streaming, fallback handling, and structured response parsing for reliable interaction.

View Project

Document RAG and Automated Resume Parser

Python, Docling, Vector RAG, Regex, NLP

Document Intelligence and Recruitment Automation

Integrated layout aware PDF extraction for tables, multiple columns, and unstructured CVs. Built vector retrieval for semantic candidate search, then used regex and NLP processing to convert extracted content into validated structured JSON.

View Project

Live Camera OCR System

Python, OpenCV, Tesseract OCR, Computer Vision

Real Time Text Digitization

Engineered a live camera pipeline for text capture and digitization. Applied dynamic regions of interest, adaptive thresholding, noise filtering, contour detection, and asynchronous frame processing to improve contrast and maintain responsive performance.

View Project

Spam Mail Classifier

Python, Scikit Learn, TF IDF, Machine Learning

Text Classification

Built a machine learning classifier for spam filtering using TF IDF features. Developed a repeatable text preprocessing workflow with normalization, tokenization, stop word removal, and model evaluation.

View Project

Research and Publications

Tiny LLMs for Efficient Bangla Cyberbullying Analysis via Knowledge Distillation and Quantization

In Preparation or Under Review

  • Developed compact language models with 10 to 60 million parameters for fine grained Bangla cyberbullying and threat detection using stratified Mendeley datasets.
  • Implemented knowledge distillation with soft targets, logit matching, and hidden layer matching from teacher models, combined with LoRA based fine tuning.
  • Evaluated 8 bit dynamic quantization on low resource hardware to improve CPU inference while preserving macro F1 performance, including attention to rare threat and harassment classes.

Design and Implementation of an Integrated RFID Based Campus Wide Payment Architecture

Undergraduate Thesis Paper and Preprint, July 2024 to December 2024

  • Engineered an embedded IoT payment ecosystem using Raspberry Pi and RFID modules for real time university transactions.
  • Designed low latency transaction and reporting logic, local caching, relational data structures, balance synchronization, and automated audit logging.
  • Evaluated concurrency, data transmission reliability, and system behavior under high density campus usage scenarios.
View Paper

Training and Certifications

Quality Assurance and Software Testing

September 2024 to December 2024

EDGE Project

  • Gained practical experience in software testing, quality assurance, bug reporting, and test execution.
  • Worked with JMeter, Selenium, Postman, and related testing workflows.

Data Science Certification Programme

March 2024 to August 2024

Ostad

  • Studied Python, SQL, statistics, data analysis, and visualization with Pandas, NumPy, Matplotlib, and Seaborn.
  • Completed practical assignments involving data cleaning, exploratory analysis, and machine learning fundamentals.

Job Readiness Training

July 2024 to August 2024

BRAC Career Hub

  • Participated in professional development sessions focused on resume preparation, interview readiness, communication, and industry alignment.

Volunteering Experience

Campus Ambassador, Farhana's Brainstation

March 2024 to December 2024
  • 🌟 Promoted training programmes and communicated learning opportunities to university students.
  • 📢 Encouraged student participation in professional development and skill building activities.

Education

April 2020 to December 2024

Bachelor of Science in Computer Science and Engineering

International Islamic University Chittagong

CGPA: 3.09 out of 4.00

Academic preparation in computer science, software engineering, data analysis, machine learning, Natural Language Processing, embedded systems, and research methodology.

References

Mr. Md. Mahiuddin

Associate Professor, Department of CSE, IIUC

Ikbal Hossen

Chief Technology Officer, Akij iBOS

Let's Connect

Discuss Research, AI Engineering, or Collaboration

I welcome conversations about Natural Language Processing, efficient language models, Bangla NLP, document intelligence, enterprise automation, and software development opportunities.

Location

Halishahar
Chattogram, Bangladesh

Phone

+8801830484860

Email

mirazchowdhury03@gmail.com

Send a Message

Complete the form and your message will be delivered directly to my email inbox.

Email Directly