Shifat IslamResearch Engineer

Shifat Islam

Research Engineer

Connecting language, vision, and multimodal AI
to real-world challenges, especially in healthcare.

I work at the intersection of natural language processing, computer vision, and multimodal AI, with a focus on Bangla, low-resource language technologies, domain adaptation, and vision–language learning. My research spans the development of language and multimodal models, datasets, and intelligent systems for real-world applications, with particular emphasis on healthcare and AI for health in low-resource settings.

Seeking PhD opportunities

Language models, multimodal learning, and AI for healthcare.

EmailGoogle Scholar ↗GitHub ↗ORCID ↗LinkedIn ↗
12Published papers
3Open datasets with DOIs
80+Citations ↗
20+Undergraduate students mentored

Research

Three interconnected research directions that shape my PhD goals.

NLP & Low-Resource Language AI

I work on language technologies for Bangla and linguistically diverse, low-resource settings, including regional language understanding, named entity recognition, hate-speech detection, healthcare language processing, and code generation from natural-language instructions. I am particularly interested in adapting language models to domain-specific and resource-constrained environments.

  • Large language models & domain adaptation
  • Bangla NLP, regional dialects & NER
  • Hate-speech detection & language understanding
  • Natural-language-to-code generation
  • Low-resource language modeling
Explore language research →

Computer Vision & Multimodal AI

I work on the intersection of language and visual understanding, developing and evaluating models and datasets for visual question answering, image classification, and vision–language learning. My work includes Bengali VQA, transformer-based image recognition, and multimodal applications in healthcare and other real-world domains.

  • Vision–language models & visual question answering
  • Bengali VQA & multimodal dataset development
  • Transformer-based image classification
  • Multimodal AI for healthcare
  • Computer vision for safety-related applications
Explore the datasets →

AI for Healthcare & Public Health

My healthcare research applies NLP, multimodal AI, and machine learning to health information and risk assessment, particularly in low-resource and LMIC contexts. My work includes AI-based risk assessment for sexual, reproductive, and mental health (SRMH), Bangla healthcare language processing, healthcare paraphrasing, and conversational AI for health information.

  • Healthcare LLMs & conversational AI
  • AI-based risk assessment for SRMH
  • Bangla healthcare NLP & paraphrasing
  • Multimodal AI for healthcare
  • AI for health information accessibility in low-resource settings
Explore healthcare projects →

Selected publications

Six representative studies across my research areas.

2026Journal

ANCHOLIK-NER: A Benchmark Dataset for Bangla Regional Named Entity Recognition.

Bidyarthi Paul, Faika Fairuj Preotee, Shuvashis Sarker, Shamim Rahim Refat, Shifat Islam, Tashreef Muhammad, Mohammad Ashraful Hoque, Shahriar Manzoor

PLOS ONE

Language & LLMsJournal
PaperarXivCodeDataset
2025Journal

Artificial intelligence-based risk assessment tools for sexual, reproductive and mental health: a systematic review.

Shifat Islam, Rifat Shahriyar, Abhishek Agarwala, Marzia Zaman, Shamim Ahamed, Rifat Rahman, Moinul H. Chowdhury, Farhana Sarker, Khondaker A. Mamun

BMC Medical Informatics and Decision Making

Healthcare AIJournal
Paper
2025Conference

Code_Gen at BLP-2025 Task 1: Enhancing Bangla Hate Speech Detection with Transformers through Token-Aware Adversarial Contrastive Training and Layer-wise Learning Rate Decay.

Shifat Islam, Abhishek Agarwala, Emon Ghosh

IJCNLP-AACL BLP

Language & LLMsConference
PaperCode
2025Conference

Comparative Study of LLMs and Transformers for Bangla Healthcare Paraphrasing.

Shifat Islam, MD. Shorif Uddin, Rayhan Uddin Bhuiyan, Azizul Fayaz, Faika Fairuj Preotee, Mohammad Shahmidul Islam

ICCIT 2025

Healthcare AIConference
Paper
2024Conference

Transformer-Based Extreme Selfie Detection for Enhancing Safety in Hazardous Environments.

Shifat Islam and Mithila Arman

IEEE WIECON-ECE

Vision & multimodal AIConference
PaperDataset
2022Conference

A Deep Learning-Based Bengali Visual Question Answering System.

M. H. Rafi, Shifat Islam, S. M. Hasan Imtiaz Labib, S. S. Hasan, F. M. Shah and S. Ahmed

IEEE ICCIT

Vision & multimodal AIConference
PaperDataset
View all 12 publicationsShow selected publications only

More publications & manuscripts

6 additional entries

2025Conference

Code_Gen at BLP-2025 Task 2: BanglaCode: A Cross-lingual Benchmark for Code Generation with Translation and Assertion Strategies.

Abhishek Agarwala, Shifat Islam, Emon Ghosh

IJCNLP-AACL BLP

Language & LLMsConference
PaperCode
2024Conference

Deep Learning-based Multi-Class Brain Abnormality Diagnosis on Bangladeshi MRI Images.

M. R. Hasan, N. Jahan, Shifat Islam and T. Muhammad

IEEE BECITHCON

Vision & multimodal AIConference
Paper
2024Conference

An Exploratory Approach Towards Investigating and Explaining Vision Transformer and Transfer Learning for Brain Disease Detection.

S. Sarker, S. R. Refat, F. F. Preotee, Shifat Islam, T. Muhammad and M. A. Hoque

IEEE ICCIT

Vision & multimodal AIConference
Paper
2024Conference

Explainable AI-Enhanced Deep Learning for Pumpkin Leaf Disease Detection: A Comparative Analysis of CNN Architectures.

M. A. Alam Khandaker, Z. S. Raha, Shifat Islam and T. Muhammad

IEEE ICCIT

Vision & multimodal AIConference
Paper
2024Conference

An Approach Towards Identifying Bangladeshi Leaf Diseases through Transfer Learning and XAI.

F. F. Preotee, S. Sarker, S. R. Refat, T. Muhammad and Shifat Islam

IEEE ICCIT

Vision & multimodal AIConference
Paper
2026Conference

VGDNet: An Attention-Based Weighted Fusion Architecture for Automated Tomato Leaf Disease Detection with Explainable AI.

Shifat Islam, Shuvashis Sarker, Bidyarthi Paul, Mahbub Hasan and Mohammad Shohel Rana

2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence & Networking (QPAIN)

Vision & multimodal AISubmitted

Full profile on Google Scholar ↗

Code & data

Open implementations and collaborative research resources.

BLP-2025 · Task 1

Bangla hate-speech identification

Transformer-based multi-task hate-speech identification, with token-aware adversarial contrastive training and layer-wise learning-rate decay.

  • PyTorch
  • Transformers
  • Bangla NLP
CodePaper

BLP-2025 · Task 2

BanglaCode

Code generation from Bengali instructions using cross-lingual translation and assertion strategies.

  • Code generation
  • LLMs
  • Cross-lingual NLP
CodePaper

Corpus & baseline models

ANCHOLIK-NER

Collaborative resources for named entity recognition across regional Bangla varieties, accompanying the PLOS ONE paper.

  • Named entities
  • Regional dialects
  • Benchmark
CodePaperData

Open datasets

Versioned research datasets linked from my ORCID record.

Dataset records on ORCID ↗

Applied projects

From research methods to working systems.

Adolescent health chatbot

Susastho.AI

Conversational AI for adolescent sexual, reproductive, and mental health awareness.

  • Flask
  • PyTorch
  • LangChain
RepositorySRMH project code

Semantic retrieval · 2025

Intent-based search engine

Semantic product search that uses intent-aware retrieval and FAISS to connect queries with relevant results.

  • FAISS
  • LangGraph
  • Redis
  • Docker
Repository

Research Mentorship

Urban VQA: Bengali Visual Question Answering for Urban Road Understanding

Multimodal AI · Bengali VQA · Urban scene understanding · Dataset development

  • Mentored the development of an urban-road Bengali VQA dataset with approximately 5,000 images and 25,000 question–answer pairs.
  • Focused on urban scenes relevant to Bangladesh and grounded visual understanding in Bangla.

Political Hate Speech Detection in the Bangladeshi Context

Bangla NLP · Hate-speech detection · Large-scale dataset development · Political text

  • Mentored the collection and curation of more than 1.5 million political text samples from multiple sources.
  • Supervised cleaning and dataset construction for Bangla political discourse and hate-speech research.

Benchmarking Multimodal Architectures for Urban VQA

Multimodal learning · Architecture benchmarking · Bengali VQA · Model evaluation

  • Mentored benchmarking of multimodal architectures for urban Bengali VQA.
  • Guided visual–textual fusion experiments and model evaluation, with attention to computational practicality.

Bangladeshi Spice Dataset, Classification & Visual Question Answering

Computer vision · Dataset development · Image classification · VQA

  • Mentored a three-stage project covering manual collection of approximately 30,000 spice images, classification, and Bengali VQA.
  • Benchmarked deep learning models for spice classification; targeting approximately 100,000 Bengali question–answer pairs.

Balanced Eye Disease Dataset & Model Benchmarking

Medical computer vision · Dataset development · Image classification · Model benchmarking

  • Mentored preparation and balancing of an eye disease dataset for medical image classification.
  • Guided deep learning experiments and architecture benchmarking to support consistent model evaluation.

Grants & research support

Research MentorREG-NER

Bangla Regional Named Entity Recognition Corpus

Institute of Research and Training, Southeast University

Research on regional Bangla named entities, with a shared corpus and baseline evaluations supporting low-resource NLP.

Grant reference: SEU/IRT/RG/2025/01/09

Research outputsANCHOLIK-NER · PLOS ONE, 2026 ↗Public dataset · Mendeley Data ↗

Project grant awarded to M. A. Hoque; my project role is Research Mentor.

Publication supportIAR · UIU

Institute for Advanced Research Publication Grant

United International University

Experience & leadership

September 2023 – Present

Healthcare AI

Research Engineer

Advanced Intelligent Multidisciplinary System Lab (AIMS Lab), IRIIC, United International University

Project: Risk Assessment Tool for Adolescent-Focused Conversational Agent in Bengali.

Developing a Bengali-language risk assessment tool and integrating it with Susastho.ai for adolescent SRMH information.

Supervisor: Prof. Dr. Khondaker Abdullah-Al-Mamun

July 2022 – August 2023

Language AI

PG Research Assistant

Bangladesh University of Engineering and Technology (BUET)

Research on named entity recognition and regional dialect modelling for low-resource languages.

Thesis: LLM Chatbot for Risk Assessment and Prediction from Textual Data.

Supervisor: Prof. Dr. Rifat Shahriyar

January 2019 – December 2021

Multimodal AI

UG Research Student

Faisal Muhammad Shah’s Lab, Ahsanullah University of Science and Technology (AUST)

Thesis: Visual Question Answering in Bengali: A Deep Learning Approach.

Publication: A Deep Learning-Based Bengali Visual Question Answering System.

Supervisor: Mr. Faisal Muhammad Shah

Education

April 2022 – Present

Postgraduate

M.Sc. in Computer Science and Engineering

Bangladesh University of Engineering and Technology (BUET)

Dhaka, Bangladesh

Thesis defence pending.

Supervisor: Prof. Dr. Rifat Shahriyar

April 2017 – January 2022

Undergraduate

B.Sc. in Computer Science and Engineering

Ahsanullah University of Science and Technology (AUST)

Dhaka, Bangladesh

Supervisor: Mr. Faisal Muhammad Shah

Awards & recognition

Best Shared Task Paper Award

BLP Workshop · IJCNLP-AACL 2025

Certificate ↗

1st / 2nd / 3rd in BLP-2025 Task 1

Subtasks 1A, 1B, and 1C, respectively

Leaderboard ↗

8th in AI Engineering Hackathon

Poridhi × BrainStation · 2025

Event ↗

Finalist

Harvard Health Systems Innovation Hackathon

Research toolkit

Language & learning

  • Python
  • PyTorch
  • TensorFlow
  • Hugging Face
  • scikit-learn
  • Vision–language models

Data & interpretation

  • pandas
  • NumPy
  • OpenCV
  • SHAP
  • Grad-CAM

Research & development

  • Jupyter
  • Google Colab
  • LaTeX
  • Git
  • GitHub

PHD OPPORTUNITIES & RESEARCH COLLABORATION

Get in touch.

I’m seeking PhD opportunities in low-resource NLP, multimodal learning, and AI for healthcare. If your group works on language technologies, multimodal AI, or their applications in healthcare and low-resource settings, I’d be glad to connect.

shifat.islam.buet@gmail.comLinkedIn ↗Google Scholar ↗A.R.I.S.E ↗