About Me
My name is Mohammad Sadegh Majidi Yazdi, but I prefer to be called Sadegh. I am currently a third-year PhD student in the Department of Computer Science at Purdue University. My current research focus is on the field of computer security, specifically security and privacy of LLMs (Large Language Models). I’ve got my B.Sc. degree in Computer Engineering at Sharif University of Technology (March 2024). My research interests are:
- Computer Security
- Large Language Models
- Computer Architecture
- Distributed Systems
As a computer scientist, I have always enjoyed studying theoretical computer science and getting my hands dirty with mathematics, but I have a special fondness for the practical aspects of computer science. I love to spend consecutive hours trying to understand a challenge in the system security and solve it.
Publications
Leaky Language Models: Stealing Architecture and Inference Optimizations via Per-Token Timing
Preprint, July 2026
project pageSadegh Majidi, Niloofar Mireshghallah, Kazem Taram
From Building Blocks to Planning: Multi-Step Spatial Reasoning in LLMs with Reinforcement Learning
Preprint, December 2025
Amir Tahmasbi, Sadegh Majidi, Kazem Taram, Aniket Bera
WIDESim: A Toolkit for Simulating Resource Management Techniques Of Scientific Workflows in Distributed Environments with Graph Topology
Journal of Grid Computing, August 2024
publication link
Mohammad Amin Rayej, Hajar Siar, Ahmadreza Hamzei, Mohammad Sadegh Majidi Yazdi, Parsa Mohammadian, Mohammad Izadi
Research Experience
Purdue University
Aug 2024 - Present
I am currently working as a research assistant under the supervision of Prof. Kazem Taram and as a full-time member of the SecArch Lab. Our lab’s general focus is on the intersection of Computer Architecture and Security, but for now, I am focusing on the security and privacy of LLMs (Large Language Models), specifically investigating potential vulnerabilities related to existing side channels in inference process of these models. As the demand for inference is growing far faster than training, we should consider the security risks of improving inference performance (latency & throughput) in our research. Understanding these risks is crucial for designing more secure and efficient AI systems.
I worked for almost a year as a research assistant under the guidance of Dr. Mohammad Izadi and as a member of the Distributed and Multi-agent Systems Lab. We worked on WIDESim, a simulation tool for resource management and scheduling of workflows in distributed computing environments with graph topology. This tool was built upon the well-known CloudSim simulator. I had the opportunity to work on various aspects of this project, including implementing different parts of the simulator, and writing a paper for publication.
Education
Purdue University
PhD in Computer Science
August 2024 - Present
Advised by Prof. Kazem Taram.
Sharif University of Technology
Bachelor of Science in Computer Engineering
September 2019 - March 2024
Graduated with a GPA of 19.19/20.
Industry Experience
Yektanet is the largest online advertising company in Iran. As a backend software engineer, I designed and developed online affiliate marketing platform (Chavosh) and product retargeting services using Django, Typescript, Postgresql and redis.
As a full-stack software engineer, I developed a new chat system (backend & frontend) for webinars using Django, React, and Lua and I also reduced live streaming latency by properly using HLS protocol over Nginx.
Selected Projects
Designed and Implemented a messaging application using Python as part of the Data & Network Security course. To satisfy a wide range of security measurements, we used a simplified version of the Double Ratchet algorithm and public-private key encryption and concepts like hashing and signing. This project is also well-structured in regard to software development standards.
This project is part of the Hardware Lab course, featuring a UEFI application designed to display weather information and forecasts for users. The primary focus lies in the ability to interact with a network interface card (NIC) within the UEFI environment and initiate a simple HTTP request using the tools provided by EDK II, the standard UEFI implementation. This project serves as an excellent learning opportunity for working with the fundamentals of the UEFI environment and stands as a valuable reference for those seeking to utilize network interfaces within this setting.
Implemented a Compiler for CMinus Langauge, a simplified subset of the C language, using Python and its standard libraries. The compiler consisted of Lexer, Parser, Code Generator, and Semantic Analyzer.
A Little More About Me
Alongside my interests in Computer Security, Computer Architecture, and Distributed Systems, some of my other interests and hobbies are:
- Participating in CTF (Capture The Flag) competitions
- Gaming, with a particular fondness for Role-Playing Games
- Swimming, my favorite sport
- Listening to music, especially in the Rock and Metal genre