Md. Raiyanul Islam

Research Interests

Artificial Intelligence
Computer Vision
Human-Computer Interaction (HCI)
Natural Language Processing (NLP)
Deep Learning

Undergraduate Research

Jan 2022 – Feb 2023

Khulna University of Engineering & Technology (KUET), Khulna, Bangladesh

Thesis: Phrase Level Bangla Sign Language Recognition using Key points from Hand Gesture Video

Developed a real-time hand gesture recognition system for Bangla Sign Language (BSL), enabling translation of sign language phrases from video input.
Implemented an LSTM model to accurately classify hand gestures from real-time video feeds.
Collaborated in constructing a novel dataset consisting of over one thousand hand gesture videos representing 27 phrases selected from the Bangla Sign Language Dictionary.
Validated the robustness of the model by achieving a 90.98% recognition accuracy on the Pakistani Sign Language dataset, surpassing the original researchers' reported accuracy of 82.66% and demonstrating the system's potential for cross-linguistic sign language recognition.
Implemented an angle based key point extraction technique to detect hand features reducing computational complexity while preserving gesture recognition accuracy using MediaPipe.

Academic Mentorship

Mar 2025 – Aug 2023

Department of Computer Science and Engineering, Dhaka International University, Dhaka, Bangladesh

Topic: A Deep Learning Approach to Predict and Monitor Student Engagement in Academic Environments

Led a group of students to design a weak-supervision pipeline to automatically generate a large labeled dataset for real-time classroom engagement detection, leveraging YOLO-based pose detection, Face Mesh, and heuristic rules using EAR, MAR, gaze vectors, and cosine similarity.
Supervised training of a YOLOv11m engagement classifier predicting five distinct classes directly from raw video frames, achieving detection accuracy of 73% (~0.73 mAP@0.5)

Independent Machine Learning Researcher

Sep 2025 – Present

Topic: Spatiotemporal Deep Learning approach for Real-Time Pedestrian Gesture Recognition

Independently formulated and executed a research initiative to solve pedestrian intent ambiguity in autonomous driving systems.
Developed a novel framework capable of processing multi-person video streams in real-time by integrating pose estimation (MediaPipe) with temporal modeling (GRU).
Optimizing model robustness by investigating Spatiotemporal Graph Convolutional Networks (ST-GCN), aiming to resolve occlusion challenges inherent in complex urban environments.

Work Experience

Software Engineer (R&D)

Sep 2025 – Present

WellDev Bangladesh Ltd.

Developed RESTful APIs with FastAPI using asynchronous features, dependency injection, Pydantic validation, and automatic OpenAPI documentation to improve backend reliability and maintainability.
Built a backend pipeline for real-time pedestrian gesture recognition by integrating MediaPipe pose extraction, GRU/GNN-based inference modules, and PostgreSQL data storage through SQLAlchemy.
Implemented essential CRUD operations, managed database migrations, and maintained data consistency within an asynchronous environment.
Containerized services using Docker for smoother cloud deployment and ensured consistent runtime environments across development and production.
Added unit and integration tests with pytest and set up a simple CI/CD-friendly workflow to support ongoing improvements to model endpoints and system performance.

Lecturer (Department of Computer Science and Engineering)

Oct 2023 – Aug 2025 (1 year 10 months)

Dhaka International University, Dhaka, Bangladesh

Delivered lectures, comprehensively assessed student learning, and provided constructive feedback.
Utilized a variety of instructional techniques, including interactive discussions, multimedia presentations, and hands-on activities, to cater to diverse learning styles and enhance student engagement.
Offered individualized academic support to native and international students foster deeper understanding of course content through dedicated mentorship and attentive guidance.
Supervised and mentored two final-year project groups (5 students each), guiding them through research, development, and implementation phases, resulting in successful project completion.
Leveraged Outcome-Based Education (OBE) for providing lectures and evaluating student performance to ensure direct alignment of course learning outcomes (CLOs) with program learning outcomes (PLOs).
Actively participated in the review and refinement of course materials and assessment and graded exams.
Streamlined the grade input process by applying an automated solution on the result management system.
Served as a counselor for two batches of over 60 students, providing comprehensive academic and personal guidance to support their successful academic journeys.

Trainee Machine Learning Engineer (Remote)

Aug 2023 – Oct 2023

Bista Solutions Ltd., Dhaka, Bangladesh

Conducted research on building a pipeline to predict sales using popular demand forecasting models like ARIMA, SARIMA, Prophet and exponential smoothing on extensive historical data.
Fine-tuned OpenAI GPT-3.5-turbo model to improve the efficiency of the NLQ to SQL conversion by 14%.
Curated a high-quality dataset of 900 NLQ-to-SQL query pairs and validated the fine-tuned model's performance.

Publications

1. Phrase Level Bangla Sign Language Recognition using Keypoints from Hand Gesture Video

International Conference on Next-Generation Computing, IoT and Machine Learning (NCIM), Gazipur, Bangladesh, 2023

Many people around the world are incapable of speaking. Sign language is the only means they use for conducting conversations. Phrase level sign language is the efficient way for hand gesture-based sign language communication. Hand detection and tracking is the crucial part for recognition of hand gesture-based sign. In research works conducted for hand gesture-based sign recognition, skin color segmentation, Kinect sensor-based data, specialized colored gloves for each hand palms etc. have been used. These research works had limitations. In this work, a dataset for recognition of 27 Bangla sign phrases was created. MediaPipe [1] library of Google was used to detect and track hands and extract key points from palm portion of each hand. Angle features between each key points of the hand palm were calculated. Long Short-Term Memory (LSTM) model was used for classification. Overall accuracy for Bangla sign language recognition by the proposed system was 92.07%. The system was also experimented on PkSLMNM [2] dataset, which is a Pakistan phrase level sign language dataset having 7 different sign phrases. Overall accuracy for recognition of phrases from PkSLMNM dataset by the proposed system was 90.98%, which was higher than the overall accuracy score of 82.66% [3] proposed by the authors of the PkSLMNM. Though the proposed system was tested on phrase level sign language recognition, it can be used in recognition of any hand gesture based sign and Human Computer Interaction (HCI).

DOI: https://doi.org/10.1109/NCIM59001.2023.10212460

S. M. M. M. Mahin, M. R. Islam and S. M. M. Ahsan, "Phrase Level Bangla Sign Language Recognition using Keypoints from Hand Gesture Video," 2023 International Conference on Next-Generation Computing, IoT and Machine Learning (NCIM), Gazipur, Bangladesh, 2023, pp. 1-6

Technical Skills

Frameworks: OpenCV, FastAPI, OpenAI, Tensorflow, Pytorch, Scikitlearn, LangChain, Keras
Database: MySQL, PostgreSQL, Firebase, SQLite
Programming language: Python, Java, C, C++
Tools & IDEs: Git, Docker, Postman, VS Code, Ollama, Jira
Documentation: Microsoft Office
Comunication: Google Workspace, Slack

Projects

Generative 3D & Face Augmented Reality

Dec 2025 – Dec 2025

Tools: Python, OpenAI Shap-E, MindAR, A-Frame, WebGL

Developed a dual-pipeline system combining Generative AI and WebAR to create interactive 3D face filters without specialized hardware.
Implemented a Python-based generation module using OpenAI's Shap-E to autonomously convert standard 2D images into textured 3D .glb models.
Built a real-time browser-based AR interface using MindAR and A-Frame that overlays generated assets onto facial landmarks with low-latency tracking and dynamic keyboard controls.

GitHub

PyRepair: A Self Correcting Agent

Mar 2025 – Mar 2025

Tools: Python, Chainlit, Google Gemini

Engineered a self-healing AI framework that autonomously detects and patches runtime errors by integrating an iterative execution-feedback loop, allowing the agent to refine broken code in real-time.
Achieved 97% problem-resolution rate (vs. 75% baseline) in a controlled ablation study, demonstrating that agentic self-correction successfully resolves complex logic failures where zero-shot models fail.

GitHub

ChatSQL

Aug 2023 – Oct 2023

Tools: Python, Langchain, LLM, OpenAI, PostgreSQL

Developed an application focusing on personalized database querying through natural language incorporating context history in the conversation to generate and process accurate SQL queries.
Optimized the performance of NLQ-to-SQL conversion by fine-tuning the OpenAI GPT-3.5-turbo model, addressing real-world database querying challenges for production level databases.
Curated a high-quality dataset of 900 NLQ-to-SQL query pairs and validated model performance.

GitHub

FaceRecog

Jan 2021 – Jun 2021

Tools: Python, sqlite, face-recognition, dlib, PyQt5

Engineered an end-to-end real-time facial recognition system for automated attendance tracking.
Utilized a secured authorization pipeline to prevent proxy attendance using eye blink detection approach.
Implemented key features including automatic attendance marking, manual attendance marking, and comprehensive data management for faculties and students of an entire institutional organization.

GitHub

Tee Store

Dec 2022 – Jan 2023

Tools: Java, Spring Boot, MySQL

Developed the back-end components of a scalable e-commerce web application using REST architecture.
Implemented the concepts of objects, classes and their relationships based with the help of Spring Boot.

GitHub

iFit

Jul 2022 – Aug 2022

Tools: Swift, Firebase

Collaborated on building the frontend portion of an iOS mobile application using Swift to provide users with structured daily exercise and workout routines.

GitHub

Awards & Certifications

Best Paper Award

International Conference on Next-Generation Computing, IoT and Machine Learning (NCIM), 2023

Awarded for the publication "Phrase Level Bangla Sign Language Recognition using Keypoints from Hand Gesture Video."
Received a 'Special Mention from Reviewers', highlighting the paper's exceptional quality and significant contribution to the field.
The proposed model demonstrated superior performance by achieving 90.98% accuracy on a public benchmark dataset, outperforming the original authors' results by over 8%.
Presented at an IEEE co-sponsored international conference, a leading forum for advancements in computing and machine learning.

[View Publication on IEEE Xplore]

Artificial Intelligence Fundamentals

IBM SkillsBuild, Issued Sep 2023

Covered core AI concepts including Machine Learning, Deep Learning, NLP, and Computer Vision.
Gained practical experience with AI model implementation using IBM Watson Studio.
Validated knowledge through a series of graded assessments and practical simulations.

[Verify on Credly]

Professional Training

Training on attaining OBE, CLOs and PLOs following guidelines of Bloom's Taxonomy

Dec 2023

Department of Computer Science and Engineering, Dhaka International University, Dhaka, Bangladesh

Strategically assessed student learning across all cognitive domains to evaluate proper CLO attainment.

Industrial attachment

Nov 2022

weDevs, Dhaka, Bangladesh

Reviewed codes and provided feedback to senior developers following agile methodologies.

Education

Khulna University of Engineering and Technology

Bachelor's in Computer Science and Engineering

Jan 2018 – Feb 2023 | CGPA: 3.64

Notre Dame College

Higher Secondary Certificate, Science

2015 – 2017 | Grade: 5.00

Ideal School and College

Secondary School Certificate, Science

Graduated: 2015 | Grade: 5.00