About Me

I'm an AI Researcher at Huawei Technologies Canada based in Edmonton, where I work on efficient on-device LLM inference: deploying and optimizing models on mobile NPUs and CPUs, and building Mixture-of-Experts inference techniques such as expert caching and prefetching and speculative decoding. During my Masters in Computing Science at the University of Alberta, I focused on neurosymbolic programming and reinforcement learning, supervised by Levi Lelis and Sandra Zilles. Before that, I completed my B.Sc. in Computer Engineering at Amirkabir University of Technology.

Beyond AI, I have a strong interest in macroeconomics and financial markets. I share my economics writing (in Persian) on my Telegram channel, @UndertowChannel, and develop tools for options trading.

Education

M.Sc. Computing Science

University of Alberta, Edmonton, AB | 2023 – 2025

Supervisor: Levi Lelis and Sandra Zilles | GPA: 3.8

Thesis Reevaluating Out-of-Distribution Generalization of Programmatic and Neural Policies through Improved Benchmarking

Thesis

B.Sc. Computer Engineering (AI)

Amirkabir University of Technology, Tehran, Iran | 2018 – 2023

Major: AI; Minor: Networks | GPA: 18.75/20 (3.96/4)

Publications

Conference Publications

Workshops

Preprints

Professional Experience

AI Researcher | Huawei Technologies Canada

Oct 2025 – Present

• Optimize and deploy large language models on mobile NPUs and CPUs, focusing on compilation, sparsification, and hardware-aware performance tuning (LiteRT Framework).
• Design expert cache eviction policies and expert prefetching strategies for memory-efficient Mixture-of-Experts (MoE) inference on resource-constrained devices.
• Develop speculative decoding methods for MoE models to accelerate token generation.
• Contributed to GLASS, a training-free test-time sparsification method for accelerating LLM inference.
• Deployed and configured Slurm-based GPU clusters, including partition setup, fair-share scheduling, and resource allocation policies for large-scale ML experiments.

Research Assistant | University of Alberta

May 2024 – Jul 2025 | Supervised by Levi Lelis & Sandra Zilles

• Refined neural policies (RL agents) through targeted architectural and training adjustments, enabling them to match or outperform programmatic policies in generalization on key benchmarks.
• Proposed new benchmarks highlighting strengths of programmatic policies (e.g., memory-based reasoning).
• Built end-to-end training/evaluation pipelines (Python, Bash, PyTorch, TensorFlow).
• Used Large Language Models to iteratively generate and refine interpretable policies in settings where RL underperforms.
• Work accepted at RLC 2025 Programmatic RL workshop and ICML 2026.

Technical Advisement | Alberta Machine Intelligence Institute (Amii)

Sep – Oct 2024

1-month project on advanced AI/ML methods to enhance real-time intrusion detection and anomaly detection in high-speed network storage infrastructures.

Machine Learning Engineer | Crouse PJS Co.

Oct 2021 – Jan 2022

• Localizing LEDs on monitors for quality control.
• Clustering LED light pixels using fuzzy C-means for luminance/wavelength extraction.
• Applying regression models per color to detect malfunctioning hardware.

Projects

I also maintain a variety of open-source projects ranging from Reinforcement Learning implementations to Computer Vision platforms.

Technical Skills

Teaching Assistant

Graduate Teaching Experience | University of Alberta

• CMPUT 365: Introduction to Reinforcement Learning
  Fall 2025 | Instructor: Marlos C. Machado

  Labs, grading assignments, and exams.

• CMPUT 366: Search & Planning in AI
  Fall 2024, Winter 2025 | Instructor: Levi Lelis

  Lectures, labs, and grading for all course components.

• CMPUT 291: Introduction to File & Database Management
  Fall 2023, Winter 2024 | Instructors: Davood Rafiei & Arash Dargahi Nobari

  Lectures and lab sessions, grading projects, assignments, and exams.

Undergraduate Teaching Experience | Amirkabir University of Technology

• Principles of AI
  Fall 2022 | Instructor: Mahdi Javanmardi

  Designed and graded projects covering CSPs, search, and RL.

• Cloud Computing
  Fall 2022 | Instructor: S. Ahmad Javadi

  Designed assignments and labs focused on APIs, cloud services, Docker, and Kubernetes.

• Internet of Things
  Fall 2022 | Instructor: Siavash Khorsandi

  Developed scenarios for sensor/actuator integration and Arduino programming assignments.

• Algorithm Design
  Winter 2021, Fall 2021, Winter 2022 | Instructors: Alireza Bagheri & Sajad Shirali-Shahreza

  Graded assignments on greedy algorithms, divide & conquer, and dynamic programming.

Honors & Awards

• Fully funded admission to UAlberta (CS, ECE, Radiology & Diagnostic Imaging), Western Ontario (CS) (2023).
• Nationwide Mathematics University Entrance Exam (Iran): Top 1% among 140,000+ applicants.
• 4-year scholarship, Amirkabir University of Technology.