Hello, I'm
Mst. RokeyaKhatun
CSE graduate from AIUB (CGPA 3.82) specializing in machine learning, deep learning, and data science. Six published ML projects on GitHub and a research paper submitted to DICTA 2026.
Mst. Rokeya Khatun
Evidence first, models second.
I'm a Computer Science & Engineering graduate from AIUB (CGPA 3.82) whose work lives in machine learning, deep learning, and data science. After completing a Data Science & Machine Learning program, I built and published six end-to-end ML projects, from a 90.5% accurate emotion classifier to a CNN served through a live Streamlit app. I've co-authored a few-shot learning paper submitted to DICTA 2026, and my undergraduate thesis pushed fine-tuned GPT models to 86.7% accuracy on IMDB sentiment analysis.
By training I'm a software engineer with strong fundamentals in OOP, data structures, and system design, and I've built the same recruitment platform in three different stacks to prove it. Machine learning and deep learning are where my real interest lives, so that's where I've pointed my career, and I'm currently learning modern AI.
Open to machine learning, data science, and research roles, remote or in Dhaka.
3.82
CGPA in BSc CSE, AIUB
6
End-to-end ML projects
2
Research works, 1 under review
86.7%
Thesis accuracy on IMDB
Selected projects
Six machine-learning builds and a three-stack recruiting platform. Every number is measured in the notebooks and reproducible from the linked repositories.
Research
One paper under review at DICTA 2026 and an undergraduate thesis on fine-tuning GPT foundation models.
Training-Free Whitened Local-Global Fusion for Few-Shot Plant Disease Recognition
Submitted to DICTA 2026 · Under review · 2nd Author
Supervisor: Md. Ismail Hossen, Griffith University, AustraliaLinkedInGoogle Scholar
Engineered a fully training-free framework on a frozen ResNet-18, combining ZCA-whitened local descriptors with a global Mahalanobis metric to bridge the laboratory-to-field domain gap in few-shot plant disease recognition.
Training-free
no fine-tuning required
Advancing NLP Tasks Through Fine-Tuning Foundation of GPT Models
Undergraduate thesis · AIUB
Developed adaptive few-shot learning techniques on top of GPT foundation models, improving the efficiency of the fine-tuning pipeline and reaching 86.7% accuracy on IMDB sentiment analysis with small labelled corpora.
86.7%
accuracy on IMDB sentiment
Technical skills
The ML stack leads: every tool here has shipped in a published project. Web development is a working foundation, not the headline. Currently learning modern AI.
ML & Data Science
Concepts
Languages
Web (working knowledge)
Databases
Tools
Experience & Education
Experience
The Premier Bank PLC
Jun 2024 - Sep 2024Intern · Gulshan Branch, Dhaka
- Completed a three-month internship embedded in the Gulshan branch's daily operations.
- Worked on the bank's challan (payment processing) workflow alongside senior staff.
- Hands-on with the bank's official software supporting real day-to-day banking operations.
Education
American International University-Bangladesh
2020 - 2025BSc in Computer Science & Engineering
CGPA 3.82 / 4.00
Thesis on fine-tuning GPT foundation models
Cambrian School and College
2019Higher Secondary Certificate (HSC)
Police Line's School and College, Kushtia
2017Secondary School Certificate (SSC)
Get in touch
Available for machine learning, data science, and research roles, plus collaborations on problems worth measuring. Based in Dhaka, Bangladesh, happy to work remote across timezones.