Portrait of Parham Zilouchian Moghaddam

Parham Zilouchian Moghaddam

AI Researcher

School of Electrical and Computer Engineering, University of Tehran

Open to Ph.D. positions and research collaborations

Curriculum Vitae Email

I study how to evaluate AI systems and how to hold them accountable.

  • Trustworthy AI
  • LLM Evaluation & Reliability
  • Efficient ML Systems
  • Generative & Multimodal Models
  • Human–AI Interaction
  • Accountable AI Agents
Now
Researcher University of Tehran
Previously
Researcher, QCRI Fanar platform, 2025–26
Education
M.Sc., University of Tehran Ranked 1st in program

I am a researcher at the University of Tehran. Since May 2026 I have been working with Dr. Mehrdad Tajbakhsh on accountability for autonomous agents. We are developing ways for an agent to record, explain, and defend its actions, and studying sovereign AI: how institutional control over AI infrastructure affects whether agents can be governed. As part of the same project, I am building an open-weight language model that assists experts with intangible-asset valuation. It draws on domain knowledge through retrieval and is designed to protect the information users share with it.

Before that, I spent a year as a researcher at the Qatar Computing Research Institute, working on Fanar, an Arabic-centric multimodal generative AI platform. I led the benchmarking of its culturally aligned image generation. I built the evaluation suite, designed fifteen LLM-as-a-judge metrics for cultural accuracy and image quality that the team used in model selection, and measured how well the automated judges agreed with Elo rankings from human pairwise comparisons. A manuscript on this work, Fanar Oryx, is in preparation [6].

I earned my M.Sc. in Computer Engineering at the University of Tehran in 2023 and graduated first among the Computer Architecture students. My thesis, advised by Dr. Mehdi Modarressi and Dr. Mohammad Amin Sadeghi, introduced NU-Class Net, a network that restores the quality of low-bitrate compressed video after standard decoding, with no change to the codec. It was published in Engineering Applications of Artificial Intelligence in 2025 [1].

I still work with two labs in Tehran. In the Computer Vision Lab, with Dr. Mostafa Tavassolipour, I am developing a generative tool that proposes carpet patterns for designers to refine, with algorithms that keep every proposal within the palette and pixel-grid limits of machine weaving. In the Intelligent Architectures for Computing Systems Lab, with Dr. Modarressi, I work on compact deep learning models for EEG analysis on wearable devices [5] and coordinate the lab’s reading group on LLM hardware and efficiency.

I have also been a teaching assistant for eight university courses, five of them as head TA, including Trustworthy AI at the University of Tehran since 2023. I was admitted to the Ph.D. program in Computer Science at UC Irvine for Fall 2024 and again for Fall 2025, but could not enroll because of U.S. visa delays. I am now looking for a Ph.D. position where I can continue this research.

Research directions

  1. Evaluating AI systems and holding agents accountable

    Methods for judging generated content, including LLM judges compared against human preferences, and mechanisms that let autonomous agents record, explain, and defend their decisions.

    Related: [6] and my current project

  2. Generative models with cultural and physical constraints

    Image generation aligned with Arabic-Islamic visual culture, and carpet-pattern generation in which every design must fit a limited palette and the pixel grid of a weaving machine.

    Related: [6] and the Computer Vision Lab project

  3. Efficient deep learning

    Getting more out of limited computation: restoring compressed video without changing the codec, compact EEG models for wearable devices, and deep learning acceleration in 3D-stacked memory.

    Related: [1] [5] [2] [7]

News

  1. New preprint on reducing the complexity of deep learning models for EEG analysis on wearable devices: arXiv:2606.12742.

  2. Started working with Dr. Mehrdad Tajbakhsh on trustworthy LLM systems and accountability for autonomous agents.

  3. Finished my year at QCRI. The Fanar Oryx manuscript on culturally aligned image generation is in preparation.

  4. Received the HATEF Technology Development Grant from the University of Tehran, chosen from more than 300 proposals.

  5. NU-Class Net published in Engineering Applications of Artificial Intelligence, vol. 144 [1].

  6. Joined QCRI as a researcher on the Fanar platform, leading the benchmarking of culturally aligned image generation.

  7. Joined the Computer Vision Lab at the University of Tehran to work on generative design for machine weaving.

  8. Smart Memory published in IEEE Computer Architecture Letters, vol. 23, no. 1 [2].

Publications

Google Scholar

Peer-reviewed

  1. [1] EAAI2025

    A novel deep learning-based approach for video quality enhancement

    P. Z. Moghaddam, M. Modarressi, M. A. Sadeghi

    Engineering Applications of Artificial Intelligence, vol. 144, art. 110118, 2025 · Introduces NU-Class Net

    DOI Code 
    @article{zilouchianmoghaddam2025video,
  author  = {Zilouchian Moghaddam, Parham and Modarressi, Mehdi and Sadeghi, Mohammad Amin},
  title   = {A novel deep learning-based approach for video quality enhancement},
  journal = {Engineering Applications of Artificial Intelligence},
  volume  = {144},
  pages   = {110118},
  year    = {2025},
  doi     = {10.1016/j.engappai.2025.110118}
}
  2. [2] IEEE CAL2024

    Smart Memory: Deep Learning Acceleration in 3D-Stacked Memories

    S. H. S. A. Rezaei, P. Z. Moghaddam, M. Modarressi

    IEEE Computer Architecture Letters, vol. 23, no. 1, pp. 137–141, 2024

    DOI 
    @article{rezaei2024smartmemory,
  author  = {Rezaei, Seyyed Hossein SeyyedAghaei and Moghaddam, Parham Zilouchian and Modarressi, Mehdi},
  title   = {Smart Memory: Deep Learning Acceleration in {3D}-Stacked Memories},
  journal = {IEEE Computer Architecture Letters},
  volume  = {23},
  number  = {1},
  pages   = {137--141},
  year    = {2024},
  doi     = {10.1109/LCA.2023.3287976}
}
  3. [3] Conference2019

    Deep Learning Based on Support Vector Machines (SVM)

    P. Z. Moghaddam, J. S. Sartakhti

    5th National Conference on Distributed Computing and Big Data Processing, 2019 · In Persian

    Paper 
    @inproceedings{zilouchianmoghaddam2019svm,
  author    = {Zilouchian Moghaddam, Parham and Sartakhti, Javad Salimi},
  title     = {Deep Learning Based on Support Vector Machines ({SVM})},
  booktitle = {5th National Conference on Distributed Computing and Big Data Processing},
  year      = {2019},
  note      = {In Persian}
}
  4. [4] Conference2019

    Improving Robot Behavior in a Real-Time Game Using the Grey Wolf Optimizer

    P. Z. Moghaddam, F. Lotfali, J. S. Sartakhti

    2nd International Computer Games Conference: Opportunities and Challenges, 2019 · In Persian

    Code 
    @inproceedings{zilouchianmoghaddam2019gwo,
  author    = {Zilouchian Moghaddam, Parham and Lotfali, F. and Sartakhti, Javad Salimi},
  title     = {Improving Robot Behavior in a Real-Time Game Using the Grey Wolf Optimizer},
  booktitle = {2nd International Computer Games Conference: Opportunities and Challenges},
  year      = {2019},
  note      = {In Persian}
}

Preprints & manuscripts in preparation

  1. [5] arXiv2026

    Reducing the Complexity of Deep Learning Models for EEG Analysis on Wearable Devices

    F. Shayeste Roodi, P. Z. Moghaddam, M. Mohammadi-Nasab, M. Modarressi, M. E. Salehi Nasab, M. Daneshtalab

    Preprint, arXiv:2606.12742, 2026

    arXiv PDF 
    @misc{shayesteroodi2026eeg,
  author        = {Shayeste Roodi, Farough and Zilouchian Moghaddam, Parham and Mohammadi-nasab, Mahdi and Modarressi, Mehdi and Salehi Nasab, Mostafa Ersali and Daneshtalab, Masoud},
  title         = {Reducing the Complexity of Deep Learning Models for {EEG} Analysis on Wearable Devices},
  year          = {2026},
  eprint        = {2606.12742},
  archivePrefix = {arXiv},
  primaryClass  = {cs.AI}
}
  2. [6] In prep.2026

    Fanar Oryx: Culturally Aligned Image Generation for Arabic-Islamic Visual Culture

    P. Z. Moghaddam, K. Isufaj, M. A. Sadeghi, R. Abolhasani, A. A. Al-Nuaimi

    Manuscript in preparation

  3. [7] In prep.2026

    Faster Wav2Vec: Improving the Inference Speed of Speech Representation Models

    P. Z. Moghaddam, M. Charbaghi, J. S. Sartakhti

    Manuscript in preparation

Research Experience

  1. May 2026 – Present

    Trustworthy LLM Systems and Agent Accountability

    Researcher · in collaboration with Dr. Mehrdad Tajbakhsh

    • Developing accountability mechanisms for autonomous agents: recording, explaining, and defending their actions.
    • Studying sovereign AI: how institutional control over AI infrastructure shapes whether agents can be governed.
    • Building an open-weight LLM co-expert for intangible-asset valuation, grounded in domain knowledge through retrieval-augmented generation.
    • Developing methods for report reliability and data anonymization to protect information users disclose.

  2. Mar 2025 – Mar 2026

    Researcher

    Qatar Computing Research Institute (QCRI) · Doha, Qatar (hybrid)

    Supervised by Dr. Mohammad Amin Sadeghi and Dr. Anas A. Al-Nuaimi

    • Led culturally aligned image-generation benchmarking for Fanar, an Arabic-centric multimodal generative AI platform.
    • Built the platform’s image-generation evaluation suite (FID, CLIP Score, Inception Score, PSNR, SSIM).
    • Designed 15 custom LLM-as-a-judge metrics for cultural accuracy and generation quality, used to inform model selection.
    • Assessed agreement between automated judge scores and Elo rankings derived from human pairwise comparisons.
    • Resulting manuscript: Fanar Oryx: Culturally Aligned Image Generation for Arabic-Islamic Visual Culture [6].

  3. Apr 2024 – Present

    Researcher

    Computer Vision Lab, University of Tehran

    Supervised by Dr. Mostafa Tavassolipour

    • Developing a generative–algorithmic assistant that proposes carpet patterns and components for designers to refine.
    • Combining generative proposals with algorithmic enforcement of limited-palette and pixel-grid constraints for machine weaving.

  4. Jan 2020 – Mar 2023

    Research Assistant

    Computational Modeling and Machine Learning Lab, University of Tehran

    Supervised by Dr. Mehdi Modarressi and Dr. Mohammad Amin Sadeghi

    • Developed NU-Class Net, a custom neural architecture for restoring visual quality in low-bitrate compressed video.
    • Used residual prediction to reduce compression artifacts after conventional decoding, without modifying the underlying codec.
    • First author of the resulting article in Engineering Applications of Artificial Intelligence (2025) [1].

  5. Nov 2019 – Present

    Research Assistant

    Intelligent Architectures for Computing Systems Lab, University of Tehran

    Supervised by Dr. Mehdi Modarressi

    • Coordinating the lab’s reading group on LLM hardware and computational efficiency.
    • Developing efficient LSTM models for EEG analysis on wearable devices, focusing on model complexity and predictive accuracy [5].
    • Co-authored Smart Memory, published in IEEE Computer Architecture Letters [2].

  6. 2018 – 2019

    Artificial Intelligence Researcher

    University of Kashan

    Supervised by Dr. Javad Salimi Sartakhti

    • Developed a hybrid DNN–SVM classifier for the B.Sc. thesis, leading to a conference publication in 2019 [3].

  7. 2017 – 2019

    Augmented Reality Researcher

    University of Kashan

    Supervised by Dr. Javad Salimi Sartakhti

    • Developed a Unity-based augmented-reality application for surface detection and placement of 3D objects.

Education & Honors

  1. Sep 2019 – Feb 2023

    M.Sc. in Computer Engineering (Computer Architecture)

    University of Tehran · Tehran, Iran

    GPA
    19.18 / 20 · highest overall GPA in the program
    Thesis
    Designing Generative Models for Developing Deep Learning Applications in Video Compression
    Advisors
    Dr. Mehdi Modarressi and Dr. Mohammad Amin Sadeghi

  2. Sep 2015 – Sep 2019

    B.Sc. in Computer Engineering (Software Engineering)

    University of Kashan · Kashan, Iran

    GPA
    17.68 / 20 overall · 18.25 / 20 in the last two years · ranked 2nd by GPA
    Thesis
    A Hybrid Deep Learning–SVM Classification Framework

Honors & awards

  • 2026

    HATEF Technology Development Grant, University of Tehran; selected from over 300 proposals.

  • 2024 – 2025

    Admitted to the Ph.D. program in Computer Science at the University of California, Irvine, for Fall 2024 and Fall 2025; unable to enroll due to U.S. visa processing delays.

  • 2023 – 2024

    Offered entrance-exam-exempt Ph.D. admission to the University of Tehran through the Exceptional Talent scheme.

  • 2023

    Graduated 1st (highest overall GPA) among M.Sc. Computer Engineering (Computer Architecture) students, University of Tehran.

  • 2019

    Direct (entrance-exam-exempt) M.Sc. admission to the University of Tehran as an Exceptional Talent.

  • 2019

    Direct M.Sc. admission offers from the University of Kashan and Shahid Beheshti University as an Exceptional Talent.

  • 2019

    Ranked 2nd by GPA among Computer Software Engineering students, University of Kashan.

  • 2018

    Recognized as a Distinguished Student by the University Committee for Exceptional Talents, University of Kashan.

Teaching & Service

Teaching assistant for eight university courses, five of them as head TA.

University of Tehran

  • 2023 – Present
    Trustworthy AI Head TA

    with Dr. Mostafa Tavassolipour

    Teach the AI ethics section; write and grade the assignments and exams; edit the lecture videos and manage the course’s YouTube channel (since its launch).

  • 2022 – 2023
    Generative Models TA

    with Dr. Mohammad Amin Sadeghi and Dr. Mostafa Tavassolipour

    Ran lab sessions on GANs, VAEs, and early diffusion models; graded assignments and final projects; edited the lecture videos for the course channel.

  • 2021, 2022
    Data Analysis TA

    with Dr. Mohammad Amin Sadeghi and Dr. Abolghasemi

    Held weekly office hours, graded coursework on statistical inference and exploratory data analysis, and gave feedback on student projects, starting with the course’s first offering in 2021.

  • 2021
    Neural Networks & Deep Learning TA

    with Dr. Ahmad Kalhor

    Graded homework and exams on backpropagation, CNNs, and RNNs for the department’s core deep learning course.

University of Kashan

  • 2022
    Machine Learning Head TA

    with Dr. Javad Salimi Sartakhti

    Wrote problem sets and led review sessions on supervised learning, SVMs, and ensemble methods.

  • 2021
    Discrete Mathematics Head TA

    with Dr. Ebrahimpour

    Led weekly problem-solving sessions and wrote the exam bank for combinatorics, graph theory, and logic.

  • 2020
    Artificial Intelligence Head TA

    with Dr. Ebrahimpour

    Directed the grading team and designed assignments on search, constraint satisfaction, and classical planning.

  • 2020
    MATLAB Laboratory Head TA

    with Dr. Ebrahimpour

    Built the lab curriculum for numerical computing and signal-processing exercises.

Earlier, I taught Java programming at Khayatzade High School (2017), covering programming fundamentals and object-oriented design (course code).

Service & outreach

  • 2021 – Present

    Writer on Medium, covering computer science and deep learning.

  • 2020 – Present

    Creator of a YouTube channel explaining computer science topics.

  • 2017 – 2018

    Executive member, University of Kashan Computer Engineering Association (CESS).

  • 2016 – 2018

    Executive member, Kashan’s Introduction to Neuroscience Workshop.

Industry Experience & Skills

  1. 2015 – Present

    Software Engineer & Board Member

    Molavi Carpet

    • Integrated multimodal generative AI into workflows for product imagery and marketing videos.
    • Lead the corporate web-platform team and administer Linux servers and network infrastructure.

  2. Jul 2024 – Feb 2025

    Computer Vision Developer

    ModAI

    • Developed object-detection and instance-segmentation models using YOLOv8 and SAM-based methods for fashion-product catalogs.
    • Designed an image-compression pipeline that reduced image payload size by 75%.

  3. 2017 – 2020

    Founder & Developer

    TechnoFarsh · carpet-industry marketplace

    • Founded a carpet-industry marketplace, incubated at the University of Kashan Innovation and Commercialization Center.
    • Built a Unity-based augmented-reality application for virtual eyewear try-on and in-room carpet visualization.

Technical skills

LLMs & agents
Open-weight model deployment (Qwen3.8, FP8), LLM APIs (OpenRouter), RAG (LangChain), LLM-as-a-judge design, prompting strategies
ML & systems
Model compression (pruning, quantization), GPU/HPC job scheduling (Slurm), generative-model metrics (FID, CLIP Score, IS, PSNR, SSIM), FFmpeg pipelines
MLOps & infrastructure
MLOps/LLMOps, Docker, Google Cloud Platform, Linux server administration, Git
Web & deployment
Django, FastAPI, Next.js, React, Node.js
Programming
Python (PyTorch, NumPy, Pandas), C/C++, SQL, MATLAB; also Julia, R, VHDL, Verilog
Languages
English (IELTS Academic 7.5), Persian (native)

Selected Projects

All repositories

  • Denoising Diffusion Probabilistic Models

    2024

    Implemented a DDPM from scratch for image denoising and generation, studying the forward and reverse diffusion processes and how variance scheduling affects sample quality.

    Diffusion · Generative models · PyTorch

  • Semantic Segmentation with U-Net

    2023

    Developed a U-Net model for pixel-level segmentation of road, vehicle, and pedestrian classes on a self-driving-car dataset.

    Segmentation · Autonomous driving

  • Car Detection with YOLO

    2022

    Built a YOLO-based detection pipeline for street imagery, producing bounding-box predictions on a real-world vehicle dataset.

    Object detection · YOLO

  • LSTM-Based Music Generator

    2022

    Trained a multi-layer LSTM for multi-genre music generation, exploring Delta-RNN and quantization for inference efficiency.

    Sequence models · Quantization

  • Bio-Inspired Computing: Optimization Algorithms

    2021

    Implemented Genetic Algorithm, Particle Swarm Optimization, K-Means, and Ant Colony Optimization for classical problems including ZOE and Vehicle Routing.

    Metaheuristics · Optimization · Code: PSO, ACO for vehicle routing

  • Parkinson’s Disease Classifier

    2021

    Designed and benchmarked 15 classifiers (KNN, DNN, SVM, ensembles) on a clinical dataset with more than 700 features, reaching 98.3% accuracy.

    Healthcare ML · Benchmarking

  • Generative Adversarial Networks (GAN) Suite

    2021

    Built and compared DCGAN, C-GAN, SR-GAN, and WGAN architectures, analyzing their convergence and output fidelity.

    GANs · Generative models

Schools, Workshops & Certificates

Summer schools, workshops & seminars

Online courses7

Contact

Email is the best way to reach me. I am glad to hear about Ph.D. openings, possible collaborations, or a paper you think I should read.

Address
School of Electrical and Computer Engineering
University College of Engineering, University of Tehran
Tehran, Iran
References
Listed in my CV, with contact details