CV

Machine Learning Specialist at Pacefactory

Pacefactory inc., Oakville, Canada

• Built a real-time object recognition system that identifies objects from motion geometry alone, without processing any image pixels. Reaches 1M inferences in 3s on standard hardware at a 0.92 macro-average score.
• {"Designed a two-tier visual state classification system for low-latency inference"=>"classical feature matching (HOG, SSIM, template matching) for high-speed scenarios, and HDBSCAN clustering over pretrained CLIP embeddings for label-free state discovery."}
• Built a self-serve tool so operators annotate object trails and train their own per-camera classifier without ML expertise.
• Accelerated annotation with Grounding DINO and Rex-Omni for text- and visual-prompted labeling, and SAM for copy-paste annotation transfer.
• Implemented iterative model-assisted annotation using YOLO and XGBoost in auto-label, correct, retrain loops.
• Developed training and export pipelines for detection, keypoint, and segmentation models, from raw video to deployment-ready outputs.

Research Assistant

University of Ottawa, Ottawa, Canada

• {"Developed a two-phase adversarial attack on deep embedding models"=>"gradient-free embedding-space search plus gradient-based input realization, crafting inputs that collectively impersonate many enrolled identities (arXiv)."}
• Developed a novel algorithm to determine whether a GAN is sufficiently trained without human oversight, validated on imagery and tabular data (published).
• Investigated generative adversarial networks for class imbalance and small-sample learning, leading to two publications.

Software Engineer at Armeh Sazeh Co.

Armeh Sazeh Co., Tehran, Iran

• Designed and implemented web-based software to automate the company's processes.
• Served as the main point of contact between the team and the client, covering communication, task delegation, requirements gathering, and back-end development.
• Designed and implemented Windows-based software in C++ to automate internal company processes.

Master of Computer Science in Applied AI

University of Ottawa, Ottawa, Canada

• Thesis - Benevolent and Malevolent Adversaries, A Study of GANs and Face Verification Systems
• GPA 10/10

Bachelor of Science in Computer Science

Amirkabir University of Technology, Tehran, Iran

• Thesis - Generating 2D and 3D human stick figure poses using Generative Adversarial Networks(GANs)
• Final Two Years' GPA 9.54/10

• Awarded "excellent student" at the University of Ottawa's graduate studies Excellence Ceremony

• Recipient of Vector Scholarship in Artificial Intelligence - $17,500

• Nazari, Ehsan, Paula Branco, and Guy-Vincent Jourdan. "Generalized Attacks on Face Verification Systems" (arXiv preprint).

• Nazari, Ehsan, Paula Branco, and Guy-Vincent Jourdan. "AutoGAN - An Automated Human-Out-of-the-Loop Approach for Training Generative Adversarial Networks." MDPI 11.4 (2023) - 977.

• Nazari, Ehsan, Paula Branco, and Guy-Vincent Jourdan. "Using cgan to deal with class imbalance and small sample size in cybersecurity problems." 2021 18th International Conference on Privacy, Security and Trust (PST). IEEE, 2021.

• Nazari, Ehsan, and Paula Branco. "On oversampling via generative adversarial networks under different data difficulty factors." Third international workshop on learning with imbalanced domains - Theory and applications. PMLR, 2021.

Papers&Snacks

Event Creator and Organizer

• Established and organized a lecture oriented weekly event for Computer Science graduate students

Running Club

Co-Event Organizer

• Co-lead a team of 15+ runners for weekly 5K sessions in Ottawa

UofO's ISAUO club

Volunteer

• Tour leader to welcome new students at the University of Ottawa

Mentorship

Volunteer

• Co-building communities for new students in Ottawa, overseeing groups of 200+ members each