2026
Evaluating Large Vision Language Models on Bangla Medical Visual Question Answering
Ahmed*, Tahmid*, Hossain*, Tomal, Jawad, Uddin, Fahim, Bhuiyan · Findings of ACL, 2026
ACL FindingsVision-LanguageMedical NLPAI Engineer & Researcher
Making large language models worth trusting.
I research how to evaluate large language and vision-language models where errors carry real cost, and how they can keep improving at test time from verifiable rewards. Applying for PhD & funded MS programs, Fall 2027.
- Peer‑reviewed papers
- 4
- As co‑first author
- 3
- Findings · 2026
- ACL
Dhaka, BD
he/him
Research focus
High-stakes evaluation of LLMs and vision-language models, test-time reinforcement learning, and parameter-efficient adaptation.
Currently
AI Engineer and Researcher at Penta Global Ltd., shipping agentic LLM pipelines for telecom while researching large language models and reinforcement learning.
Looking for
PhD & funded MS in NLP, ML, computer vision, or LLM & network security, starting Fall 2027.
News
Oct 2026
RxFact, my first-author benchmark of clinically harmful errors in prescription images, submitted to ACL Rolling Review. Details →
Apr 2026
Our evaluation of large vision-language models on Bangla medical VQA accepted to Findings of ACL 2026. Paper ↗
Dec 2025
BanglaMedVQA dataset and benchmark accepted to the 2nd AAAI Bridge Program on AI for Medicine and Healthcare. Paper ↗
Nov 2025
Two papers, LP-FT-LoRA and our PentaML shared-task system, accepted to the BLP Workshop at IJCNLP-AACL 2025.
Nov 2025
Placed top-15 in all three subtasks of BLP-2025 Shared Task 1 (Bangla hate speech detection).
Recent publications
Published
Read ↗
2026
How Good LLMs Are at Answering Bangla Medical Visual Questions? Dataset and Benchmarking
Ahmed*, Tahmid*, Hossain, Tomal, Fahim, Bhuiyan · AAAI Bridge Program, 2026
VQAMedical NLPBangla2025
LP-FT-LoRA: A Three-Stage PEFT Framework for Efficient Domain Adaptation in Bangla NLP Tasks
Tomal*, Uddin*, Tahmid, Hossain, Fahim, Bhuiyan · BLP Workshop (IJCNLP), 2025
PEFTLoRABangla NLP2025
PentaML at BLP-2025 Task 1: Linear Probing of Pre-trained Transformer-based Models for Bangla Hate Speech Detection
Tahmid*, Ahmed*, Jawad, Uddin, Fahim, Bhuiyan · BLP Workshop (IJCNLP), 2025
Hate speechLinear probingBanglaUnder review
2026
RxFact: Can Large Language Models Detect Clinically Harmful Errors in Prescription Images?
Tahmid, Mazumder, Hossain, Sakib, Shahriar, Reshma, Fahim, Islam, Bhuiyan · Under review, ACL Rolling Review (Oct 2026)
Vision-LanguageMedical NLPPatient safety2026
AD-SFL: Activation-Space Anomaly Detection and Reinforcement Learning-Based Defense Against Data Poisoning in SplitFed Learning
Tomal*, Tahmid*, Roy, Hasan, Razzaque · Under review, Computer Networks (Elsevier)
SplitFed learningData poisoningSecurity*Equal contribution.
See all publications and statement of research →
Mentors & affiliations
Faculty advisor
Dr. Md. Abdur Razzaque
Professor & Chairman, Dept. of CSE
University of Dhaka
Green Networking Research Group
Research mentor
Md Fahim
Doctoral Student
The University of Texas at Dallas
Formerly Penta Global Data Lab
Industry supervisor
Md Farhad Alam Bhuiyan
Head of R&D
Penta Global Ltd
Penta Global Data Lab
© 2026 Intesar Tahmid