EXTRA
Research Motivation
My research is driven by a fundamental question: how do we build AI systems that are simultaneously powerful, accessible, and trustworthy? This question has guided my work across parameter-efficient adaptation, adversarial robustness, and equitable AI development.I am motivated by the practical reality that most of the world cannot afford massive computational resources or extensive labeled datasets. This drives my focus on efficient fine-tuning methods that maintain model performance while dramatically reducing resource requirements. The challenge of making models robust against adversarial manipulation similarly stems from a commitment to deploying reliable systemsβunderstanding failure modes is essential for building AI we can trust in critical applications. A core conviction underlying my work is that AI's benefits should not be limited by linguistic or economic barriers. My efforts in low-resource language processing aim to extend state-of-the-art capabilities to underserved communities, recognizing that true progress in AI means progress for everyone, not just well-resourced domains.
Looking forward, I am increasingly interested in machine unlearning and model editing - the ability to precisely modify what models know is crucial for addressing evolving privacy requirements, correcting biases, and maintaining model relevance over time. Ultimately, I aspire to develop adaptation and control mechanisms that make advanced AI both more capable and more responsible.
π Education
-
2023 - present PhD
Boston University, Boston, MA, USA -
2017 - 2021 BSc. in Computer Science and Engineering
Shahjalal University of Science and Technology, Sylhet, Bangladesh
πΌ Experience
-
2022 - 2023 Research Developer
Wikimedia Foundation, San Francisco, CA, USA -
2021 - 2022 Machine Learning Engineer
GigaTech Ltd., Dhaka, Bangladesh -
2022 Research Assistant
BRAC University, Dhaka, Bangladesh -
2021 - present Research Affiliate
Bengali.AI, Dhaka, Bangladesh
π€ Research Interests
-
Primary
- Parameter-efficient Finetuning, Model Editing, Adversarial Attacks, Machine Unlearning
-
Secondary
- Bias Mitigation & Fair Representations, Robust Domain Adaptation,
- Knowledge Infusion, AI For Social Good
π οΈ Skills
-
Programming Languages
- Python, C++, Java, JavaScript, SQL, SPARQL
-
Machine Learning Frameworks
- PyTorch
- Keras
- OpenCV
- pySpark
-
Frameworks
- Flask, Django, Android
-
Tools and Misc.
- Git, Docker, Linux, Bash, LaTeX, Markdown, HTML, CSS, Firebase
π§ Leadership & Service
-
Leadership
- Founding Member | SUST ACM Student Chapter
- Judge | SUST Tech Fest DL Enigma, BDOSN NLP Hackathon
- Organizing Committee | IEEE ICBSLP, 2019. SUST Tech Fest, 2020
-
Service
- Mentor - HSA , Intl. ML Olympiad
- Volunteer - Outreacy, BDOSN Ada Lovelace Celebration,
π Accolades
-
- π₯ Champion, BRACATHON 3.0 - Healthcare, 2019
- π₯ Champion, SUST Tech Fest - Hackathon, 2019
- π 4th place, AI-Based Dhaka Traffic Detection Challenge, 2020
- π» Grace Hopper Scholarship Recipient, 2021
- π₯ 2nd Runner-up, CVPR AI City Challenge Track-4, 2022
- π₯ 2nd Runner-up, Microsoft Imagine Cup SEANM, 2022
- π Outreachy Internship at the Wikimedia Foundation, 2022
- π Dean's Fellowship, Boston University, 2023
- π« CRA-WP Grad Cohort Scholarship, 2025
- π CVPR WiCV Mentorship Grant, 2025
π Non-academic Interests
-
- HOZIER π’πΈ
- Malai Chaa β
- Analog Horror π½οΈ
- Folk Horror ππΌππΌπ
- Gothic Horror π¦
- Cosmic Horror π
- Ritual Horror π―οΈ
- Ambient Horror π«οΈ
- Magical Realism β¨
- Fall of my enemies β οΈ
- Comics π¦ΈββοΈ
- Anime productions by Mappa π
- Seas with warm water π
- Anything by Neil Gaiman and Alan Moore π