I am an ML/NLP researcher and I hold a degree in Industrial and Production Engineering from Shahjalal University of Science and Technology and have spent the past four years building a research record across multimodal LLMs, computational biology, and human-AI interaction in healthcare settings.
My research centers on a single goal: building human-centric, reliable, and safe AI systems, with healthcare as a primary application domain. I pursue this across three interconnected directions:
I collaborate with Prof. Razzak (MBZUAI) and Prof. Chae (HYU) on GenAI, LLM-HCI, and biomolecular ML; Riashat Islam, PhD (Microsoft Research) and Md Rizwan Parvez, PhD (QCRI) on biomedical AI, agents, and reasoning; researchers from Cohere Labs on LLM evaluation and alignment; and Prof. Min Xu (CMU) on biomolecules. I completed HTGAA 2025 (MIT Media Lab) on protein engineering and joined as a Global TA. I also founded CIOL, a student-led research lab at SUST, where I mentor junior researchers in collaboration with Prof. Ahsan (OU).
My research has been published in venues including ICLR (A*), ICML (A*), ACL (A*), WWW (A*), EMNLP (A*), FAccT, CSCW (A), IEEE TCBB, ACCV, DASFAA, IISE, and COLING, and co-located workshops. I regularly review for major AI/ML venues (NeurIPS, ICLR, ICML, TMLR, AAAI, T-PAMI, ACL, EMNLP, EACL, CHI) and received the ICML 2026 Gold Reviewer award. I am also a Kaggle Grandmaster.
Feel free to email me (azminetoushik dot wasi at gmail dot com) or connect via the links below if you are interested in collaboration or discussing research.
multilevel-legal-reasoning dataset!
→ View my list of Rejections and Failures!
SpatiaLab: Can Vision-Language Models Perform Spatial Reasoning in the Wild?
Azmine Toushik Wasi, Wahid Faisal, Abdur Rahman, Mahfuz Ahmed Anik, et al.
Benchmarks VLMs on real-world spatial reasoning tasks, revealing systematic failures in grounded scene understanding.
ICLR 2026
[PDF] ▪
▪
Position: AI Governance Needs ISO-like Interoperability Protocols, Not Just Laws
Azmine Toushik Wasi, Mst Rafia Islam, Mahfuz Ahmed Anik, Md Manjurul Ahsan, TH Rafi, Dong-Kyu Chae
Argues for standardized machine-readable governance protocols over jurisdiction-specific legal frameworks, drawing on ISO interoperability models.
ICML 2026: Position Papers (Spotlight, Top 5%)
(A*)
TimeSpot: Benchmarking Geo-Temporal Understanding in Vision-Language Models in Real-World Settings
Azmine Toushik Wasi*, Shahriyar Zaman Ridoy*, Koushik Tonmoy, Kinga Tshering, et al.
Evaluates VLMs on geo-temporal grounding using real-world imagery, exposing gaps in temporal and location reasoning.
ICML 2026
(A*)
From Language Specifications to Executable Turing Machines: Evaluating LLMs as Computational Machine Designers
Shahriyar Zaman Ridoy, S. M. Muhtasimul Hasan, Azmine Toushik Wasi, et al.
Tests whether LLMs can translate formal language specifications into correct, executable computational machines.
EMNLP 2026 (Findings, Top 30%)
(A*)
Frugal Medical AI as an Equity Imperative: Rethinking Algorithm Design for Resource-Constrained Healthcare
Azmine Toushik Wasi, Mohsin Mahmud Topu, Mahfuz Ahmed Anik, Md Manjurul Ahsan
Proposes a compute-frugal design framework for clinical AI to reduce performance gaps in low-resource settings.
ACM EAAMO 2026 (Top 22.5%, Oral)
PD-scWorld: Pathway-Guided Disentanglement for Single-Cell Perturbation World Models
Azmine Toushik Wasi (Solo Author)
Introduces pathway-guided disentanglement for learning interpretable world models of single-cell responses to perturbations.
MLCB 2026 |
ACM BCB 2026
(Top 20%, Oral)
CADGL: Context-Aware Deep Graph Learning for Predicting Drug-Drug Interactions
Azmine Toushik Wasi, TH Rafi, Raima Islam, Serbetar Karlo, Dong-Kyu Chae
Context-aware GNN for DDI prediction, outperforming prior graph-based methods on standard benchmarks.
IEEE/ACM TCBB (Q1, IF: 4.5, CiteScore: 9.7) ▪
[DOI] ▪
[arXiv]
When SMILES have Language: Drug Classification using Text Classification Methods on Drug SMILES Strings
Azmine Toushik Wasi, Karlo Serbetar, Raima Islam, TH Rafi, Dong-Kyu Chae
Frames drug SMILES strings as natural language and applies text classification methods, achieving competitive drug categorization results.
ICLR 2024
Tiny Papers ▪
[OpenReview] ▪
[arXiv] ▪
[GitHub] ▪
Multimodal Vision-Language Models for Automated and Explainable Postoperative Complication Risk Stratification
Azmine Toushik Wasi, Mahfuz Ahmed Anik, Md Shafikul Islam, Md Manjurul Ahsan
VLM-based pipeline for predicting and explaining postoperative complication risk from multimodal clinical data.
IISE 2026 (Health Systems Track, Best Track Paper: 2nd Place) ▪
[PDF]
View Publications and Ongoing Works →
Visiting Researcher, MBZUAI | April 2026 - Present
Research Collaboration, Qatar Computing Research Institute | August 2025 - Present
Research Collaboration, Microsoft Research | January 2026 - March 2026
Global TA (Volunteer), How to Grow Almost Anything (HTGAA), MIT Media Lab | August 2025 - February 2026
Group Member, Schwartz Reisman Institute for Technology and Society (SRI), University of Toronto | October 2025 - March 2026
Visiting Researcher, AI4CHEMIA Research Group, King Saud University | April 2025 - September 2025
Fellow (School of AI), Pi School | June 2025 - August 2025
Student Research Assistant, Mila Quebec AI Institute | May 2024 - March 2025
Community Researcher, Cohere Labs | August 2024 - January 2026
Founding Researcher, Computational Intelligence and Operations Laboratory (CIOL) | March 2021 - Present
Research Intern, Xu Lab, Carnegie Mellon University | July 2024 - December 2024
Visiting Researcher, Data Intelligence Lab, Hanyang University | October 2023 - Present
Copyright © 2022-2026 Azmine Toushik Wasi. All rights reserved.