Portrait of Arash Rezaali

Arash Rezaali

Computer Engineering Student · Researcher
Professional email
arash.rez03@gmail.com
Phone number
+98 991 233 1107
GitHub
github.com/Arash-ra03
LinkedIn
linkedin.com/in/arash-rezaali
Google Scholar
Google Scholar profile
Final-year BSc Computer Engineering student at Shahid Beheshti University focused on reinforcement learning, robotics, multi-agent systems, and reasoning in large language models. I co-developed a SEMI-CTDE region-based multi-agent deep RL architecture for adaptive traffic signal control and am extending this line of work in my bachelor thesis using graph neural network–based MARL on urban traffic networks. I also have hands-on experience with modern large language models through the LLM Armenia summer school and applied NLP and vision–language projects. I enjoy tackling difficult problems, learning new things, and exploring fields in depth.
I am currently seeking Master’s and PhD positions aligned with these interests.
Publications
Semi-CTDE paper thumbnail
A Semi Centralized Training Decentralized Execution Architecture for Multi Agent Deep Reinforcement Learning in Traffic Signal Control
A. Rezaali, P. Yazdani, M. Abdoos

Proposes a transferable SEMI-CTDE region-based multi-agent RL architecture for adaptive traffic signal control with two complementary model variants and extensive ablations.

↗ View preprint on arXiv ↗ View code on GitHub
Experience
LLM Armenia logo
LLM Armenia 2025 Summer School
Armenia · Summer 2025
Selected Participant

Completed a one-week international LLM program blending theory with implementation-focused work on six projects, including ViT-from-scratch training, SigLIP+Qwen LaTeX OCR, GRPO post-training for symbolic reasoning, inference-time compute strategies, and adversarial attacks on CLIP and vision models.

↗ View LLM Armenia projects on GitHub
Shahid Beheshti University logo
Shahid Beheshti University
Tehran, Iran · 2022 – 2026
Research Assistant – Intelligent Transportation Systems Lab

Working under the supervision of Dr. Monireh Abdoos on SEMI-CTDE region-based multi-agent deep reinforcement learning architectures for adaptive traffic signal control in SUMO. This line of work has resulted in a manuscript currently under review (see Publications).

Awards
Olympiad silver medal
Silver Medal – National Astronomy & Astrophysics Olympiad
Iran · 2020

Awarded silver medal in the highly competitive national Olympiad, demonstrating strong problem-solving skills in physics and mathematics.

Education
Shahid Beheshti University logo
BSc in Computer Engineering
Shahid Beheshti University · Tehran, Iran · 2021 – 2026
GPA: 17.8 / 20 (3.79/4.00) · Top 5%
Key A+ coursework (GPA 4.0 / 4.0)
Intro to Machine Learning AI & Expert Systems Fundamentals of Robotics Algorithm Design Signals & Systems Graph Theory & Algorithms Eng. Stats & Probability Linear Algebra Software Engineering
Young Scholar Club logo
Young Scholar Club – Olympiad Summer Camp
Iran · Summer 2020

Selective Olympiad summer camp for the top 40 candidates nationwide, combining advanced astronomy and physics instruction with frequent exams and problem-solving workshops used to determine final medal rankings.

Projects
LLM Armenia project screenshot
LLM Armenia 2025 Summer School
Language Models · Deep Learning · Machine Learning

Participated in the LLM Armenia 2025 Summer School, blending theoretical concepts and practical projects in machine learning. Key projects included ViT-from-scratch training, SigLIP+Qwen LaTeX OCR, GRPO post-training for symbolic reasoning, inference-time compute strategies, and adversarial attacks on CLIP and vision models.

↗ View LLM Armenia projects on GitHub
Multi-Class Emotion Recognition using DistilBERT
Final project – Introduction to Machine Learning
Multi-Class Emotion Recognition using DistilBERT
Natural Language Processing · Emotion Recognition · Multi-Class Classification

Implemented an end-to-end emotion recognition pipeline: fine-tuned a pre-trained DistilBERT model on a custom emotion dataset, then used its embeddings as input to an MLP classifier with per-label threshold tuning to improve classification performance. The project includes dataset preprocessing, hyper-parameter search, and a simple real-time inference.

↗ View on GitHub
Robotics Course project screenshot
Fundamentals of Robotics Course – Projects
Mobile Robotics · Kinematics · Localization · Navigation

Developed a sequence of assignments and a final project for a mobile robotics course, covering kinematics and wheel-velocity control, localization and hallway mapping, local navigation with Bug algorithms, and an indoor mapping and navigation project where a Mavic platform and TurtleBot perform wall-following, door detection, and particle-filter-based localization to reach target doors.

↗ View on GitHub
Connect Four AI – minimax and Q-learning agents
Final project – Artificial Intelligence
Connect Four AI using Minimax and Q-Learning
Game AI · Minimax · Alpha-Beta Pruning · Q-Learning

Implemented two AI agents for the Connect Four board game. The first uses a minimax search with alpha–beta pruning and configurable depth to control playing strength; the second is a tabular Q-learning agent trained via large-scale self-play. The project contrasts search-based and reinforcement-learning approaches to decision making in a fully observable game.

↗ View on GitHub
Unity 2D Platformer Game Design final project
Final project – Game Design
Unity 2D Platformer
C# · Unity · 2D Platformer

Built a tiny 2D platformer in Unity with a responsive run/jump controller, simple enemies with health bars, and collectible hearts. The game features multiple themed biomes (forest and snow) with a smooth transition and a clean project structure, making it a reusable template for future sidescroller ideas.

↗ View on GitHub
Teaching & Voluntary Work
Shahid Beheshti University logo
Teaching Assistant
Shahid Beheshti University · Tehran, Iran · 2022 – 2026
Core CS and AI courses
Teaching responsibilities
  • Design quizzes for weekly sessions and exams.
  • Design programming and written (paper) assignments.
  • Receive and organize assignments and course projects.
  • Design end-of-term and mini-projects.
  • Hold problem-solving and review sessions.
  • Provide solutions and explanations to students’ questions.
  • Grade assignments, projects, and exam components.
Courses supported
Artificial Intelligence Discrete Mathematics Linear Algebra Compiler Design Signals & Systems Advanced Programming Formal Languages & Automata
Astronomy & Astrophysics Olympiad preparation class
Instructor – Astronomy & Astrophysics Olympiad Preparation
Top High Schools in Iran · 2021 – Present
Lead Instructor – Advanced Mathematics & Physics (Olympiad Level)

Teach advanced Mathematics and Physics to high-performing high school students preparing for the National Astronomy & Astrophysics Olympiad, with a strong focus on rigorous problem solving and competition strategy.

Lessons taught
Stellar Structure & Evolution Orbital Mechanics Cosmology Galaxy Dynamics Calculus Data Analysis Astrophysics Classical Mechanics

Over the years, I have had the privilege of teaching students who later achieved national and international medals in Astronomy & Astrophysics Olympiads through their own hard work and dedication.

Students solving Olympiad problems in class
Olympiad problem set on the board
× Enlarged view