About Me

Hello There!

I am an Applied ML Research Engineer with 5+ years of experience delivering large-scale Computer Vision and Deep Learning solutions for end-to-end ML systems in production.

I have a proven track record of translating research into scalable ML products, including distributed inference pipelines processing 187M+ image operations. I am also experienced in building and optimizing models across diverse data modalities for performance, robustness, and deployment.

My Journey

  • Canada

  • Machine Learning Research Engineer II

    Captura (Skylab Technologies, acquired) | Nov. 2024 – Feb. 2026 | Vancouver, BC

    End-to-end ML for portrait imaging: cropping, color correction, face landmarks — deployed at scale via ONNX + Triton.

    Show details
  • Machine Learning Research Engineer I

    Skylab Technologies Inc. | Aug. 2022 – Nov. 2024 | Vancouver, BC

    Production ML pipelines for portrait hair matting, face super-resolution, and large-scale synthetic data generation.

    Show details
  • Machine Learning Developer Intern

    Kinaxis | May 2021 – Aug. 2021 | Ottawa, ON

    ML model validation tooling and ensemble time-series forecasting for supply chain demand prediction.

    Show details
  • M.Sc. in Computing Science — GPA 4.0/4.4

    Simon Fraser University | Sept. 2019 – Dec. 2021 | Burnaby, BC

    Research Assistant, Network & Multimedia Systems Lab (Prof. Mohamed Hefeeda). Thesis published in IEEE TMM 2022.

    Show details

  • Iran

  • B.Sc. in Electrical Engineering, Minor in Computer Engineering

    University of Tehran | Sept. 2014 – Jul. 2019 | Tehran, IR

    Thesis on Deep RL for NFV service provisioning — published at IEEE GLOBECOM 2019.

    Show details
  • Undergraduate Research Assistant

    AMCS Lab – University of Tehran | May 2018 – May 2019

    DQN framework for reliability-aware NFV resource allocation.

    Show details
  • Iranian National University Entrance Exam

    Ranked 71st out of 250,000 participants | 2014

    Member of Iran's National Elites Foundation.

My Skills

Programming Languages

  • Python (Proficient)
  • C / C++ (Intermediate)
  • Git / GitHub / Linux

Deep Learning & ML

  • PyTorch / Keras / TensorFlow
  • OpenCV / NumPy / scikit-learn
  • TensorRT / ONNX / OpenVINO
  • LlamaIndex / Pandas

Cloud & DevOps

  • AWS / S3
  • Docker
  • Apache Kafka
  • SQLite / PostgreSQL
  • Triton Inference Server

Computer Vision

  • Image Segmentation / Matting
  • GAN / Super-Resolution
  • Contrastive / Self-Supervised Learning
  • Synthetic Data Generation

ML Infra & Deployment

  • DVC / MLflow
  • ONNX / OpenVINO export
  • Distributed training (DDP)
  • PyTorch Lightning / Hydra

NLP & Generative AI

  • LLM / RAG (LlamaIndex)
  • OpenAI GPT-4 / Qwen2.5-VL
  • vLLM / Chainlit
  • Multimodal Annotation Pipelines

Publications

  • Unsupervised Single-Image Reflection Removal

    IEEE Transactions on Multimedia (TMM) • 2022

    S.H. RahmaniKhezri, S. Kim, M. Hefeeda. Proposed the first unsupervised framework using cross-coupled Perceptual Deep Image Priors. Achieved 37% improvement over prior methods, 24% PSNR and 12% SSIM gains in benchmarks.

  • Deep Reinforcement Learning for Dynamic Reliability Aware NFV-Based Service Provisioning

    IEEE GLOBECOM • 2019

    S.H. RahmaniKhezri et al. DQN-based framework for reliability-aware NFV service placement optimizing admission ratio and cost under bandwidth and reliability constraints. IEEE Xplore →

Presentations

Recent presentations with the posters and reports

Dec. 6th 2019, Vancouver, Canada

Improving Visual Question Answering Using Semantic Analysis and Active Learning

In this work, we aim to train a model for the task of visual question answering, using only a small number of labeled data.

Read more

In order to do so, we used the ideas influenced by the active learning. We defined an oracle to provide a label for the question that is asked about an image. This oracle is an image captioning network, that given an image as its input, generates a sentence that describes the objects that are visible in that image. We use a semantic similarity calculator, in order to connect the result of the image captioning model and interpret that to become a potential label for the visual question answering task.

Read less

SFU AI Poster Session 2019
Project Report
Dec. 12th 2019, Waikoloa, USA

Deep Reinforcement Learning for Dynamic Reliability Aware NFV-Based Service Provisioning

In this work, we considered a dynamic reliability-aware NFV placement using DQN.

Read more

We considered a multi-InP scenario in which different levels of reliability with different costs are offered to the network operator. On the other hand, we considered multiple incoming services with different reliability requirements. For DQN-agent, we defined the state set, action set, reward and memory considering the objective of the NO which is maximizing the admission ratio while minimizing the placement cost. Using simulations, we showed that the NO could learn how to effectively use the resources of the InPs for various services in different states in a way that the admission ratio is maximized and placement cost is minimized.

Read less

IEEE GLOBECOM Conference Presentation
IEEE GLOBECOM paper