About me

I am currently pursuing a Master's degree in Artificial Intelligence at the University of Amsterdam. My research interests include Large Language Models, Vision-Language Models, Explainable & Interpretable AI, and AI Safety & Security. I actively participate in research and projects aimed at advancing trustworthy and human-centered AI systems. My Bachelor's degree in Computer Engineering equipped me with a solid engineering mindset and a strong mathematical foundation, enabling me to approach AI challenges from both theoretical and practical perspectives.

Interests

  • Large Language Models

  • Vision-Language Models

  • Natural Language Processing

  • Representation Learning

  • Explainable and Interpretability AI

  • AI Safety, Security & Alignment

  • Trustworthy AI

Resume

Education

  1. University of Amsterdam

    2025 — Present

    MSc in Artificial Intelligence.

  2. University of Guilan

    2019 — 2023

    B.Sc in Computer Engineering.
    GPA: 3.8/4

Work Experience

  1. Data Analyst At Anhar Company

    March 2025 - September 2025

    Engineered an AI-driven project management system utilizing Large Language Models (LLMs) and Retrieval Augmented Generation (RAG) to automate project control protocols. Integrated the LLM pipeline with backend databases and designed interactive dashboards, streamlining data visualization and decision-making for real-time project monitoring..

  2. Research Assistant at Guilan University

    Jan 2023 - September 2025

    I lead a research initiative supervised by my professor, focus on various medical images.

Teaching Experience

  1. Artificial Intelligence

    Fall 2022

    University of Guilan
    Instructor: Dr. Y. Boreshban
    Head TA

  2. Algorithm Design

    Fall 2021

    University of Guilan
    Instructor: Dr. A. Khozaei
    Head TA

  3. Data Structure

    Fall 2021

    University of Guilan
    Instructor: Dr. F. Feyzi
    Head TA

  4. Discrete Mathematics

    Fall 2020 / Fall 2021

    University of Guilan
    Instructor : Dr. S. M. Shekarian
    Head TA

  5. Digital Circuits

    Fall 2021

    University of Guilan
    Instructor: Dr. M. Aminian
    Head TA

  6. Computer Architecture

    Fall 2021

    University of Guilan
    Instructor: Dr. H. Ahmadifar

  7. Computer Aided Design

    Fall 2022

    University of Guilan
    Instructor: Dr. M. Aminian
    Head TA

  8. Microelectronic Circuits

    Spring 2023

    University of Guilan
    Instructor: Dr. M. Aminian

Projects

Research Papers

  1. Sparse-Attention Transformers for Efficient Whole-Slide Image Classification

    Submitted to NeurIPS 2026

    Investigated sparse-attention transformer architectures for scalable and data-efficient Whole-Slide Image (WSI) classification in computational pathology. Developed and benchmarked multiple Vision Transformer variants, including sparse and deformable attention mechanisms, against state-of-the-art Multiple Instance Learning approaches, demonstrating their effectiveness for large-scale pathology image analysis.

Selected Projects

  1. Hyperbolic Alignment for Figurative Language in Vision–Language Models

    Investigated figurative language understanding in Vision-Language Models by modeling image–text relationships as hierarchical visual-semantic structures. Fine-tuned a hyperbolic representation learning model (FigMERU) on the IRFL benchmark, demonstrating improved generalization to unseen figurative categories such as metaphors compared to Euclidean baselines.

  2. Collapsed Language Models Promote Fairness

    Conducted a reproducibility study of the ICLR 2025 paper Collapsed Language Models Promote Fairness, evaluating Uniform Neural Collapse (U-NC3) as a fairness indicator in encoder-only language models. Adapted Neural Collapse metrics for BERT-based masked language models and developed a unified experimental pipeline for reliable fairness analysis.

  3. OCR Application

    Developed a full-stack OCR web application that extracts text from uploaded images through an image-to-text processing pipeline. Built a FastAPI backend, integrated OCR processing, and containerized the system with Docker to provide a scalable and user-friendly text extraction service.

  4. Universal Dependencies LinguisticStudy

    This project investigates the effects of partially freezing layers during the fine-tuning of a distilled multilingual model (DistilBERT) using the Universal Dependencies dataset. The study focuses on the task of Part-of-Speech (PoS) tagging and aims to explore how layer freezing impacts both model performance and training efficiency in this specific linguistic analysis context.

  5. Data-Efficient Human Sperm Abnormality Detection Using Uncertainty Sampling

    led a research initiative focused on enhancing sperm abnormality detection using various transformers models like VIT and Swin. We have recently worked on using contrastive learning and active learning on this dataset.

  6. Amazon Reviews

    A machine learning model was developed and trained on Amazon Reviews to analyze user sentiment. The dataset was extensively preprocessed, involving techniques such as tokenization, stopwords removal, lemmatization, stemming, tag and emoji removal, and normalization.
    The sentiment analysis task on Amazon reviews was conducted using Logistic Regression. The repository provides resources and code for a variety of sentiment classification models, including Embedding Models, BERT, and Simple Neural Networks.

  7. An Odd Music Generator

    This project aims to explore Autoencoders Denoising, data augmentation methods, audio file processing, string-to-string models (Seq2seq), and intelligent systems.
    The project consists of four key components: Denoising, Note Recognition, Note Prediction, and Noise Maker.
    I used an LSTM-based n-gram language model to predict the next musical note in a sequence, supporting the development of well-structured musical compositions.

  8. Analyzing Human Metaphase II Oocyte Images

    This project serves as an example of semantic segmentation, featuring the implementation of a robust deep learning-based multiclass semantic segmentation method. It is specifically designed for analyzing human metaphase II oocyte images, as outlined in a paper

  9. Trie

    For my university's Data Structure course project, I developed a student management system. I utilized a HashTable to store student data indexed by their IDs and implemented a Trie tree to store HashTable's keys.

Skills

Programming Languages

Python

Java

C++

Machine Learning Frameworks

Pytorch

TensorFlow

Keras

Scikit-Learn

LangChain

Huggingface

Data Visualization

Numpy

Pandas

Matplotlib

Seaborn

Web Development

HTML

CSS

MySQL

Operating Systems

Linux(Ubuntu)

Extra Tools

Git

Latex

Certification