I am an AI researcher and engineer with a B.Sc. in Computer Science and Engineering from North South University (CGPA 3.86/4.00, 100% Merit Scholarship). I work across computer vision, multimodal learning, and NLP, with six accepted papers, three of them in Q1 journals.
My goal is to build faithful, interpretable, and deployable AI systems, along three directions:
- Computer Vision and Multimodal Learning: cross-modal alignment, modality balancing, graph-based representations, and test-time adaptation under distribution shift.
- NLP and Agentic Systems: LLM evaluation on politically sensitive text, multi-agent systems, and tool orchestration.
- Trustworthy and Explainable AI: calibration, uncertainty, and interpretability of deep models, with a focus on Medical AI, clinical decision support, and high-stakes diagnostic applications.
I collaborate with Dr. Mohammad Ali Moni (Charles Sturt) on explainable medical AI. I also recently collaborated with Prof. Alex Lamb (College of AI, Tsinghua University) on a recent ICLR submission on retrieval geometry in cache-based CLIP adaptation. Earlier I worked with Dr. Nabeel Mohammed and Dr. Shafin Rahman at the NSU Machine Intelligence Lab on multimodal learning (CAT‑GS), with an industry collaboration with RobotBulls on a real-time avatar system.
My work appears in Knowledge-Based Systems (IF 8.0), Neurocomputing (IF 6.7), PLOS ONE, AACL, and a NeurIPS workshop, with a first-authored ICLR 2027 submission under review. I was a Teaching Assistant for Design and Analysis of Algorithms (600+ students) and am a Codeforces Expert. I am currently a Research Intern at CIOL, working on visual anomaly localisation, LLM evaluation, and reliable tool-using agents, and a Software Engineer at bKash, the leading mobile financial service (MFS) provider in Bangladesh.
I am seeking PhD and M.Sc. research opportunities. If my work aligns with your group, I would be glad to hear from you at mahir.tamim@northsouth.edu.
News and Updates
- October 2026 Research Harness Recession is accepted to SLM-Agents, the 1st NeurIPS Workshop on SLMs for Agentic Systems (co-first author)!
- October 2026 Career Joined bKash as a Software Engineer to work in the Backend team.
- September 2026 Research Preprint Retrieval Geometry Shapes Cache-Based Clip Adaptation is on arXiv, under review at ICLR 2027.
- September 2026 Research Our paper on evaluating crisis sentiment in Bangladesh's July Uprising is accepted to AACL 2026 (co-first author)!
- August 2026 Career Joined SysModeler.ai as a Backend Engineer, working on an AI platform for model-based systems engineering (SysML v2).
- July 2026 Research CAT-GS is accepted for publication in Neurocomputing (Q1, IF 6.7) as first author!
- January 2026 Research Our rainfall classification framework is published in PLOS ONE (Q1, IF 2.6) as first author.
- January 2026 Career Joined Obby LLC as a Product Engineer, building agentic and voice-AI systems.
- December 2025 Academic Completed my B.Sc. in CSE at North South University — CGPA 3.86/4.00 (96.5%).
- December 2025 Academic Defended my undergraduate thesis, PNEUMA, a model-agnostic long-term video memory architecture for egocentric question answering.
- December 2025 Research Paper accepted at IEEE ICCIT 2025 on sentiment analysis during Bangladesh's 2024 mass uprising.
- October 2025 Research Our multi-teacher knowledge distillation work is published in Knowledge-Based Systems (Q1, IF 8.0).
- April 2025 Research Joined the NSU Machine Intelligence Lab and RobotBulls as a Research Assistant in computer vision.
- January 2025 Academic Served as Problem Setter for the Intra-NSU Programming Contest.
- January 2024 Academic Awarded a 100% Merit Scholarship by North South University for outstanding academic performance.
Selected Publications
All publications →-
Retrieval Geometry Shapes Cache-Based Clip Adaptation
Shows that the choice of retrieval encoder, not the memory bank itself, determines how much cache-based test-time adaptation helps CLIP, and proposes MARC, a training-free pairing of frozen CLIP predictions with DINOv2-B retrieval.
-
CAT-GS: Balanced Multimodal Learning via Calibrated Gating and Fusion Surgery
A training method that stops one input type from dominating multimodal learning, so the model makes better use of every modality without any change to its architecture.
-
Can LLMs Follow the Pulse of a Crisis? Evaluating Crisis Sentiment in Bangladesh's July Uprising
A dataset of about 200K Bangla social media comments from the 2024 uprising, labelled by native speakers, showing that even strong LLMs struggle to read public sentiment during a fast-moving crisis.
-
Multi-Teacher Knowledge Distillation and Ensemble Algorithm for Efficient Brain Tumor Classification in Resource-Constrained Environments with Explainable AI
Distils an ensemble of teacher networks into a compact student for brain tumour classification on constrained hardware, with explainability retained end to end.
Research Experience
Full experience →-
Research Intern, CIOLMay 2026 – Presentwith Dr. Dong-Kyu Chae (Hanyang University)
I lead SphereAD, a training-free way to find anomalies in images that works across domains without fitting anything per domain. I also work on small language model agents, which is where our Harness Recession workshop paper came from.
-
Research Assistant, Charles Sturt UniversityDec 2025 – May 2026
Mostly medical AI. We compressed several large teacher models into one small student for brain tumour classification (Knowledge-Based Systems). I also co-led the AACL 2026 crisis-sentiment paper, built on a dataset my friends and I put together ourselves.
-
Research Assistant, NSU Machine Intelligence Lab & RobotBullsMar 2025 – Nov 2025with Dr. Nabeel Mohammed and Dr. Shafin Rahman
Two things: CAT-GS, our multimodal learning paper now in Neurocomputing, and a real-time talking avatar for RobotBulls. Most of my time on the avatar went into lip sync and cutting latency until it could hold a live conversation.
-
Teaching Assistant & Competitive Programming Instructor, NSUJan 2024 – Dec 2025
Two years of tutorials, grading, and office hours for more than 600 students in CSE 373: Design and Analysis of Algorithms. I also taught at the NSU Problem Solvers Bootcamp and set problems for intra-NSU contests. Writing problems and their test data made me think carefully about edge cases and about explaining an algorithm so that someone else can actually use it.