experience

Collaborating with Dr. Dong-Kyu Chae, Associate Professor & PI, Data Intelligence Lab at Hanyang University and a multi-institutional research team on culturally grounded evaluation of moral reasoning and ethical alignment in large language models.

• Co-authored BengaliMoralBench: A Benchmark for Auditing Moral Reasoning in Large Language Models within Bengali Language and Culture, published at the ACM Conference on Fairness, Accountability, and Transparency (FAccT 2026).
• Contributed to the development and evaluation of BengaliMoralBench, a culturally grounded benchmark comprising 3,000 Bengali moral scenarios across five sociocultural domains, 50 subtopics, and three ethical frameworks: Virtue, Commonsense, and Justice.
• Contributed to systematic evaluation and analysis of open-weight and closed-source LLMs, examining cultural grounding, ethical reasoning, prompt-language effects, model-scale behavior, and failure patterns across culturally nuanced scenarios.

Collaborating with Dr. Md. Mosaddek Khan, Associate Professor & PI, Cognitive Agents and Interaction Lab at the University of Dhaka on formal reasoning and the evaluation of executable symbolic systems generated by large language models.

• Co-authored From Language Specifications to Executable Turing Machines: Evaluating LLMs as Computational Machine Designers, accepted at EMNLP 2026 Findings.
• Evaluated eight LLMs across 253 formal-language specifications, comprising 2,024 model–specification pairs, 1.21 million possible test evaluations, and 563,400 executed behavioral tests.
• Analyzed static, construction, and semantic failure modes, revealing an overall exact-pass rate of 11.12% and indicating that difficulty varies more with operational structure than with the traditional context-free versus non-context-free distinction.

Collaborating with Dr. Md Rizwan Parvez, Research Scientist at the Qatar Computing Research Institute (QCRI) and a multi-institutional research team on geo-spatial and geo-temporal reasoning in vision-language and large language models.

• Contributed to benchmark development and empirical evaluation of vision-language models in real-world geographic and temporal settings.
• Co-authored TimeSpot, a benchmark for geo-temporal understanding in vision-language models, accepted as a regular paper at ICML 2026.
• Conducted structured synthesis of methods, benchmarks, evaluation settings, and research gaps for Geo-Spatial and Geo-Temporal Reasoning in Vision-Language and Large Language Models: A Review, accepted to the Main conference at AACL-IJCNLP 2026.

• Assisted faculty with student advising, research activities, lecture preparation, and course support.
• Provided tutorial sessions and communicated academic updates and announcements to students.
• Checked examination scripts, prepared grade sheets, and proctored examinations.

• Assessed quizzes and coursework, including written assignments, using instructor-provided rubrics to ensure fair and consistent grading.
• Tabulated and verified scores and prepared mark sheets and assessment summaries for instructor review.
• Proctored departmental examinations while following confidentiality, communication, and academic-integrity policies.