Experience

Vanderbilt University

Cognitive Agents and Interaction Lab, University of Dhaka

Supervisor: Dr. Md Mosaddek Khan

• Investigate whether LLMs can synthesize executable formal systems (grammars, pushdown automata, Turing machines) from natural-language specifications, judged by behavioral correctness rather than surface plausibility.
• Two papers accepted to Findings of EMNLP 2026, one as first author; a third paper on context-free grammar correctness is under review at ACL Rolling Review.

Qatar Computing Research Institute (QCRI)

Supervisor: Dr. Md Rizwan Parvez

• Design evaluation protocols for geo-temporal and spatial reasoning in vision-language models under real-world uncertainty.
• Co-first author of TimeSpot (ICML 2026) and co-author of SpatiaLab (ICLR 2026); lead author of a survey on geo-spatial and geo-temporal reasoning accepted to AACL-IJCNLP 2026.

Computational Intelligence and Operations Laboratory (CIOL)

Distinguished Advisors: Prof. Dong-Kyu Chae (Hanyang University), Prof. Md Manjurul Ahsan (University of Oklahoma)

• SEISMOS (NeurIPS 2026): co-developed a method that treats semantic chunking of documents as a statistical signal-detection problem, placing chunk boundaries where a shift in meaning is statistically significant rather than at fixed lengths or hand-tuned similarity thresholds.
• InvariantBench (ICML 2026 AI4Math Workshop; extended version under review at ICLR 2027): contributed to a benchmark that tests whether LLMs reach the same answer when a problem is rewritten into a mathematically equivalent form, separating genuine reasoning from surface pattern matching.
• Retrieval geometry in CLIP adaptation (under review at ICLR 2027; with Prof. Alex Lamb): studied how the geometry of the retrieval cache governs how well cache-based methods adapt CLIP to new tasks.
• Small language models (NeurIPS 2026 SLM-Agents and MusIML workshops): examined when agent scaffolding stops helping small models as they improve, and separated instruction compliance from answer correctness in post-training evaluation.
• AI-text detectors and creativity; clinical NLP robustness (ICML 2026 Co-Creativity Workshop; Findings of AACL 2026): analyzed how AI-text detectors judge human creative writing, and argued that robust clinical NLP needs messy, multimodal, longitudinal, privacy-preserving corpora.

Apurba-NSU R&D Lab, North South University

Supervisors: Dr. Nabeel Mohammed, Dr. Shafin Rahman

• Studied knowledge distillation and other compression techniques for large language models.
• Joint first author of CAPSTONE (EMNLP 2025 Industry Track), a plug-and-play scene-to-text pipeline that lets a frozen LLM reason over visual input, matching fully trained VLMs at a fraction of the cost.

Artificial Intelligence Institute, University of South Carolina (AIISC)

Collaborators: Prof. Amitava Das (BITS Pilani, formerly AIISC), Aman Chadha (AWS), Vinija Jain (Meta AI)

• Researched hallucination detection in text-to-image generation using multimodal models such as Qwen-VL and LLaVA, evaluated on MS-COCO and synthetic T2I benchmarks.
• Explored speculative-decoding defenses against adversarial attacks on diffusion models.

North South University

Supervisor: Dr. Rashedur M. Rahman

• Built a context-aware data-cleaning pipeline for Bengali text that preserves sentence-level context while removing noise, improving transformer accuracy by up to 4% on three benchmarks (SN Computer Science, Springer Nature).
• Developed an efficient preprocessing pipeline for clinical text with transformer encoders, cutting training time by up to 17% (Best Paper Award, IEEE IS 2024).

Department of Electrical and Computer Engineering, North South University

CSE 225L Data Structures and Algorithms Lab · CSE 115L Programming Language I Lab

• Deliver coding-focused lab lectures and hands-on exercises on algorithms, data structures, and programming fundamentals.
• Lab Instructor Coordinator for CSE 225L; updated lab manuals, redesigned problem sets, and standardized lab operations across sections.
• Proctor coding assessments, debug student code, and grade assignments, viva tests, and lab exams.

Department of Electrical and Computer Engineering, North South University

CSE 225 Data Structures and Algorithms · CSE 215 Programming Language II

• Mentored 160+ undergraduates per semester in data structures, algorithms, and object-oriented programming.
• Led tutorial sessions and graded assignments with detailed, formative feedback.

NSU Problem Solvers

ICPC · Codeforces · AtCoder · Meta Hacker Cup · Vjudge

• Honorable Mention, ICPC Asia Dhaka Regional Online Preliminary.
• Solved roughly 1,000 problems across Codeforces, CodeChef, AtCoder, and LightOJ; mentored peers in data structures and algorithms.