Curriculum Vitae

Below is a concise version of my curriculum vitae. For a PDF copy please download the full CV. See also the Research page for expanded project descriptions and the Publications page for papers.

Education

University of Isfahan (UI)

B.S. in Computer Engineering • Sept 2019 – Jul 2024
Isfahan, Iran

• GPA: 3.83/4.0 (17.94/20).
• GPA (last four semesters): 4.0/4.0.
• Thesis: Graph Fusion in Multi‑modal Networks with Zero‑shot Evaluation.
• Ranked in the top 5 % of more than 120 students.

Research Interests

• Multimodal large language models (LLMs) and retrieval‑augmented generation.
• Trustworthy and interpretable AI, including uncertainty calibration and privacy preservation.
• Generative and diffusion models for data synthesis and model auditing.
• Zero‑shot and few‑shot learning in educational and medical domains.

Publications

Below are recent and forthcoming publications. For a complete listing please refer to my Publications page.

• Hidden in Plain Sight: Evaluating Abstract Shape Recognition in Vision‑Language Models – NeurIPS 2024. (with A. Davies, T. Lamb, J. Yuan, P. Torr, A. Khakzar, F. Pinto)
  Introduced IllusionBench, a dataset of images containing hidden letters, faces and animals to audit VLM perceptual robustness; humans achieve near‑perfect accuracy whereas SOTA models score below 40 % zero‑shot.
• Survey on LLM in Requirement Engineering – Frontier Journal 2024. (with M. Sharbaaf, S. Kolahdouz‑Rahimi, K. Lano)
  Comprehensive survey of how large language models are being applied to requirements engineering tasks from elicitation to validation.
• MEENA (Persian MMMU): Multimodal‑Multilingual Educational Exams for N‑level Assessment – under review (COLM 2025). (with O. Ghahroodi et al.)
  Created the first large‑scale Persian benchmark for vision‑language models with 7,500 Persian and 3,000 English multimodal questions covering scientific reasoning, problem solving and student performance metadata.
• RAG‑Driven Video QA with Adaptive Chunking – CSICC 2025. (with M. Hassan Heydari, K. Vadaei, M. Shirian, A. Fatemi)
  Leveraged retrieval‑augmented generation to answer questions about educational videos using a bilingual dataset.
• Leveraging Retrieval‑Augmented Generation for University Knowledge Retrieval – IKT 2025 (oral). (with K. Vadaei, M. Hassan Heydari, A. Fatemi)
  Proposed a two‑stage RAG pipeline with Persian LLMs for domain‑specific information retrieval.
• Context Awareness Gate for Retrieval‑Augmented Generation – IKT 2025. (with M. Hassan Heydari, E. Naman, A. Fatemi)
  Introduced the Context Awareness Gate, a mechanism that dynamically determines when to retrieve external knowledge before prompting an LLM.
• CLIP Exhibits Improved Compositional Generalisation Through Representation Disentanglement – preprint (arXiv 2024). (with R. Abbasi, M. Samiei, M. H. Rohban, M. S. Baghshah)
  Analysed the compositional generalisation of CLIP and the impact of training data diversity and representation disentanglement.
• Advanced Mutation Testing with Zero‑ and Few‑Shot Evaluation using GPT‑v4 – under review (CSICC 2025). (with F. Aghababaei, M. Sharbaaf)
  Applied GPT‑v4 to generate and evaluate mutations for software testing in low‑data regimes.

Research Experience

Video‑LLM Uncertainty Calibration and Training Data Extraction

Torr Vision Group, University of Oxford • Sept 2024 – Present
Under Prof. Philip Torr

• Evaluating and improving uncertainty calibration for video‑based VLMs.
• Generating multiple‑choice question datasets from video using diffusion models to study memorisation.

Vision‑Language Benchmark in the Educational Domain

RIML, Sharif University of Technology • Aug 2024 – Present
Under Prof. Mohammad Hossein Rohban

• Building a bilingual benchmark of multimodal multiple‑choice questions spanning elementary through high school curricula.
• Designing metadata and evaluation protocols for zero‑shot, few‑shot and hallucination detection experiments on GPT‑4, Gemini and other VLMs.

Shape vs. Scene Reasoning Benchmark

Torr Vision Group, University of Oxford • Mar 2024 – Aug 2024
Under Prof. Philip Torr

• Developed a dataset to test whether models rely on object shape or scene context; released as IllusionBench (NeurIPS 2024).
• Coordinated human subject experiments and model fine‑tuning to achieve a 24‑point accuracy gain.

3‑D Vessel Segmentation & Counterfactual Modelling

CAMP Chair, Technical University of Munich • Sep 2023 – Mar 2024
Under Prof. Nassir Navab

• Applied geometric deep learning and Fourier transforms to segment vascular structures in 3‑D images.
• Worked on counterfactual modelling using vision‑linguistic models such as DRAGON.

Robustness in Vision‑Language Models

RIML, Sharif University of Technology • Feb 2023 – Oct 2023
Under Prof. Mohammad Hossein Rohban

• Investigated disentanglement in contrastive models like CLIP and assessed out‑of‑distribution robustness.

Zero‑Shot Learning in the Medical Domain

ILS Lab, University of Isfahan • Aug 2021 – Nov 2022
Under Dr. Peyman Adibi

• Leveraged manifold learning and geometric approaches to bridge the gap between seen and unseen classes on the ChestX‑ray8 dataset.

Honours & Awards

• TUM Summer Research Fellowship, Technical University of Munich, 2023.
• Top 5 % Academic Ranking, University of Isfahan, 2023.
• 2nd place National Data Science Competition, Isfahan, 2022.
• 22nd place, ICPC Asia Tehran, 2020.
• 3rd place, Poytek Business Summer School, 2020.
• Diploma of Honour, International Kangaroo Mathematics Competition, 2015.

Teaching

I have served as a teaching assistant for a variety of undergraduate and graduate courses, including:

• Deep Learning
• Machine Learning
• Knowledge Systems
• Medical Image Processing
• Advanced Programming
• Linear Algebra
• Artificial Intelligence
• Engineering Probability & Statistics
• Computer Architecture
• Object‑Oriented Analysis & Design
• Fundamentals of Robotics and Data Structures

Leadership & Community

• Quera Data Team Course Instructor – designed and delivered courses on LLMs, deep learning and NLP (Jun – Oct 2024).
• BigData Lab, University of Isfahan – board member researching retrieval‑augmented generation (2023 – Present).
• Community for Artificial Intelligence (UI) – founder and chairman, promoting AI research and outreach (2020 – Present).
• Rasta Association (SUT) – chairman and development lead (2023 – Present).
• TEDx University of Isfahan – organiser, co‑organiser and advisor for multiple events (2020 – 2023).

Skills

• Machine Learning: PyTorch, NetworkX, scikit‑learn, OpenCV, pandas, NumPy, Matplotlib; basic familiarity with TensorFlow and Keras.
• Programming: Proficient in Python and C++; working knowledge of Java, MATLAB and VHDL.
• Languages: Native Persian; advanced English (TOEFL 93/120).