Itamar Tsayag, Developer in Tel Aviv-Yafo, Israel
Itamar is available for hire
Hire Itamar

Itamar Tsayag

Verified Expert  in Engineering

Bio

Itamar is a talented algorithm developer and data enthusiast with expertise in computer vision, machine learning, and statistical analysis. He has successfully deployed cutting-edge algorithms for enhancing IVF cycle efficacy and stroke diagnosis. With a master's degree in electrical engineering and data science, Itamar excels in articulating complex topics and driving impactful results, leveraging his technical prowess and business acumen.

Portfolio

Sanas.ai
Machine Learning, ONNX Runtime, PyTorch, OpenCL...
EIN Plus LLC dba GovPlus
Computer Vision, OpenCV, Image Processing, PyPI, Production, Apache Airflow...
ONE TO ONE SYSTEMS INC
Computer Vision, Python, Machine Vision, CAD, OpenCV, PyTorch...

Experience

  • Python - 8 years
  • PyTorch - 7 years
  • Deep Learning - 7 years
  • Data Science - 5 years
  • Algorithms - 4 years
  • Research - 4 years
  • Computer Vision Algorithms - 3 years
  • Medical Imaging - 2 years

Availability

Full-time

Preferred Environment

Ubuntu Linux, PyCharm, Visual Studio Code (VS Code), Amazon EC2, PIP, Git, Bitbucket, Vim Text Editor, MacOS, Stock Trading

The most amazing...

...experience I've had was researching and developing a neural network compression algorithm for smaller networks with comparable performance.

Work Experience

Machine Learning Engineer

2024 - 2024
Sanas.ai
  • Converted speech models to the ONNX format to enable compatibility across multiple platforms and in real-time streaming.
  • Designed a versatile software framework for converting models to an ONNX format.
  • Developed a benchmarking system for ONNX models across multiple platforms.
Technologies: Machine Learning, ONNX Runtime, PyTorch, OpenCL, Open Neural Network Exchange (ONNX), Realtime, Machine Learning Operations (MLOps), Machine Learning Algorithms, Infrastructure, Data Inference, AI Integration, Quantization, Neural Network Pruning, model quantization, Developing AI Models locally, Recurrent Neural Networks (RNNs), LSTM Networks, Data Processing, Testing

AI Developer

2024 - 2024
EIN Plus LLC dba GovPlus
  • Developed a computer vision pipeline that transformed selfie images into compliant, regulation-ready passport photos for submission to the US government.
  • Built the infrastructure for this type of algorithmic pipeline, including Python packaging, asynchronous execution with Asyncio, and latency optimization.
  • Served as the central point of expertise for all computer vision algorithm areas within the team. The algorithm successfully replaced a third-party API, leading to significant resource savings for the company.
Technologies: Computer Vision, OpenCV, Image Processing, PyPI, Production, Apache Airflow, PyTorch Lightning, Machine Learning Algorithms, Infrastructure, Data Inference, AI Consulting, Hugging Face, Llama, Algorithm Design, Developing AI Models locally, Streamlit, SQL, Image Segmentation, Pose Estimation, MediaPipe, Data Processing, Testing

AI Developer

2023 - 2024
ONE TO ONE SYSTEMS INC
  • Designed and developed an algorithm from scratch that converts body measurements into a CAD file outlining a garment pattern.
  • Integrated human body models to extract body measurements, a critical component of the company’s algorithm and pipeline.
  • Developed the algorithm to maturity, complete with a user-friendly front end, and was successfully demonstrated in a polished POC that gained positive feedback from investors.
Technologies: Computer Vision, Python, Machine Vision, CAD, OpenCV, PyTorch, Convolutional Neural Networks (CNNs), PyTorch Lightning, Machine Learning Algorithms, AI Consulting, Algorithm Design, Developing AI Models locally, Data Classification, Streamlit, Image Processing, Image Segmentation, Pose Estimation, MediaPipe, Testing

Lead Algorithm Developer

2022 - 2023
AIVF
  • Developed an SW and ML pipeline of the embryo grading algorithm used for enhancing IVF cycle efficacy.
  • Worked on an algorithm that leverages cutting-edge video classification and segmentation networks on a complex microscopy image dataset.
  • Created an algorithm that was successfully deployed as the flagship product of the company, demonstrating strong expertise in code productization.
Technologies: Videos, Python, PyTorch Lightning, Optimization, Embryology, Visual Studio Code (VS Code), Vim Text Editor, ClearML, Data Science, Data Scientist, Scikit-learn, Matplotlib, Data Extraction, Visualization, Open Neural Network Exchange (ONNX), Health, Startups, Data Analytics, Statistical Analysis, Data Reporting, Research, XLSX File Processing, Microsoft Excel, Monte Carlo Simulations, CSV File Processing, Data Cleansing, Data Structures, Data Modeling, Regression Modeling, Large Data Sets, Parallelization, Feature Engineering, Cloud, Data Engineering, Data Pipelines, Pattern Recognition, Machine Vision, API Integration, Minimum Viable Product (MVP), Model Development, Transformers, Technical Leadership, AI Modeling, AI Model Training, JMP, CSV Export, AWS Lambda, Data Inference, AI Integration, Algorithm Design, Data Classification, SQL, Image Processing, Object Detection, Image Segmentation, FastAPI

Senior Computer Vision Algorithm Developer

2020 - 2022
Viz.ai
  • Developed an algorithm to differentiate between damaged brain tissue to healthy brain tissue in cases of strokes.
  • Oversaw the entire project from annotation guidelines, database structure, system architecture, and SW development to training deep learning models.
  • Oversaw the SW quality and acted as a tech lead, including code reviews, packaging, and structure.
  • Managed the tasks of a ten-member team using Jira. The team included data managers and algorithm developers.
Technologies: Python, Deep Learning, Computer Vision Algorithms, Analytics, Databases, Software, Algorithms, PyTorch, Amazon Web Services (AWS), Convolutional Neural Networks (CNNs), TensorFlow, Keras, Artificial Intelligence (AI), Software Architecture, Jira, Slack, Machine Learning, Image Analysis, Jupyter Notebook, Classification Algorithms, APIs, Artificial Neural Networks (ANN), Neural Networks, JSON, Functional Programming, DevOps, Terraform, Decision Trees, Pytest, Unit Testing, Object-oriented Programming (OOP), Git, GitHub, Team Leadership, Medical Imaging, Ubuntu Linux, PyCharm, PIP, OpenCV, Cron, CSV, Pandas, NumPy, Data Analysis, Statistics, Jupiter, Amazon EC2, Amazon S3 (AWS S3), Docker, Back-end, Back-end Development, Linux, AI Programming, Healthcare, AWS DevOps, Codebase Development, Data Visualization, Plotly, Amazon SageMaker, Automation, JupyterLab, XGBoost, Random Forests, 2D, GPU Computing, Graphics Processing Unit (GPU), 2D Modeling, Machine Learning Operations (MLOps), Test-driven Development (TDD), Deep Neural Networks (DNNs), Computer Vision, 3D, Distributed Computing, Data Loading, Machine Learning Automation, Supervised Machine Learning, Videos, Visual Studio Code (VS Code), Vim Text Editor, ClearML, Data Scientist, Scikit-learn, Matplotlib, Data Extraction, Visualization, Open Neural Network Exchange (ONNX), Health, Startups, Data Analytics, Statistical Analysis, Data Reporting, Research, XLSX File Processing, Microsoft Excel, Object Detection, CSV File Processing, Data Cleansing, Data Modeling, Regression Modeling, Large Data Sets, Parallelization, Feature Engineering, Cloud, Data Engineering, Pattern Recognition, Numba, Python Performance, Machine Vision, API Integration, Minimum Viable Product (MVP), Model Development, Technical Leadership, Data Scraping, AI Modeling, AI Model Training, CSV Export, AWS Lambda, You Only Look Once (YOLO), Data Inference, AI Integration, Data Classification, SQL, Image Processing, Image Segmentation, Pose Estimation

Computer Vision Algorithm Engineer

2020 - 2021
Defense Companies (Classified)
  • Developed an algorithm for the real-time detection of objects in a multisensor video.
  • Augmented geographical objects as landmarks on the video images using 3D LLA to 2D coordinates projection.
  • Used the algorithm to support detection and tracking, enabling high video frames per second (FPS) and supporting multiple video domains with RGB and thermal infrared images.
Technologies: Python, Video Processing, Image Processing, Optical Sensors, Deep Learning, Computer Vision Algorithms, PyTorch, Amazon Web Services (AWS), Convolutional Neural Networks (CNNs), TensorFlow, Keras, Machine Learning, Image Analysis, Jupyter Notebook, Classification Algorithms, Artificial Intelligence (AI), APIs, Artificial Neural Networks (ANN), Neural Networks, JSON, Functional Programming, Decision Trees, Object-oriented Programming (OOP), Git, GitHub, Ubuntu Linux, PyCharm, PIP, OpenCV, Cron, CSV, Pandas, NumPy, Data Analysis, Statistics, Jupiter, Amazon EC2, Fine-tuning, Linux, AI Programming, Codebase Development, Data Visualization, Plotly, Automation, JupyterLab, XGBoost, Random Forests, 2D, GPU Computing, Graphics Processing Unit (GPU), Machine Learning Operations (MLOps), Deep Neural Networks (DNNs), Computer Vision, Analytics, Data Loading, Machine Learning Automation, Supervised Machine Learning, Visual Studio Code (VS Code), Vim Text Editor, Data Science, Data Scientist, Scikit-learn, Matplotlib, Data Extraction, Integration, Visualization, Open Neural Network Exchange (ONNX), Startups, Statistical Analysis, Optimization Algorithms, Microsoft Excel, CSV File Processing, Data Cleansing, Regression Modeling, Large Data Sets, Feature Engineering, Numba, Python Performance, Machine Vision, Model Development, Data Scraping, Physics, Optical Systems, AI Modeling, Scraping, AI Model Training, Image Segmentation

Senior R&D Engineer

2018 - 2020
Magic Leap
  • Developed the procedures and algorithms for calibrating a novel time-of-flight (ToF) depth sensor. All calibrations were factory-ready with minimum hardware requirements and high-performance demands.
  • Adapted the Eulerian video magnification (EVM) algorithm to estimate the user's heart rate from the Magic Leap headset's eye-tracking cameras. This was part of the initial research into the viability of using the headset for medical purposes.
  • Analyzed cross-platform data to enable critical architectural decisions regarding the development of the head-mounted display.
  • Developed a method for generating synthesized calibration vectors that enabled performance testing of the head-mounted displays on realistic edge-case calibration scenarios.
Technologies: Python, Sensor Fusion, Sensor Data, Optical Sensors, Simulations, Computer Vision Algorithms, Statistical Methods, Calibration, Deep Learning, Architecture, Hardware, Machine Learning, Computer Vision, PyTorch, Convolutional Neural Networks (CNNs), Depth Sensors, Image Analysis, Jupyter Notebook, Artificial Neural Networks (ANN), Neural Networks, JSON, Decision Trees, Object-oriented Programming (OOP), Git, GitHub, Ubuntu Linux, PyCharm, PIP, OpenCV, CSV, Pandas, NumPy, Data Analysis, Statistics, Jupiter, Linux, AI Programming, Data Visualization, Plotly, Monte Carlo Simulations, JupyterLab, Random Forests, GPU Computing, Graphics Processing Unit (GPU), Deep Neural Networks (DNNs), Analytics, Facial Recognition, Data Loading, FFmpeg, Supervised Machine Learning, Visual Studio Code (VS Code), Vim Text Editor, Data Science, 3D Pose Estimation, Data Scientist, Matplotlib, Integration, Startups, Statistical Analysis, Optimization Algorithms, Microsoft Excel, CSV File Processing, Data Cleansing, Regression Modeling, Large Data Sets, Feature Engineering, Machine Vision, Data Scraping, Digital Elevation Models, GIS, Physics, Optical Systems, Electrical Engineering, Virtual Reality (VR), Augmented Reality (AR)

Experience

Facial Landmark Detection Using Visual Transformers

Researched the use of Visual Transformers for facial landmark detection. The goal was to improve the performance on faces with occlusions, which was previously hindered by the heatmap regression method. The transformers improved results on occluded landmarks through self-attention but resulted in an increased error when handling images of faces without occlusions.

Uncovering a Winning Lottery Ticket with Stochastic Gates

A novel pruning technique based on stochastic gates was developed. By leveraging this approach on over-parametrized neural networks, a subnetwork capable of achieving comparable results to the target network was discovered without any additional training. This research offers a promising avenue for reducing computational costs and memory requirements while maintaining performance, with practical benefits across various domains utilizing neural networks.

EyeRate: Estimating Heart BPM Using AR HMD

EyeRate was a self-directed project that utilized the eye-tracking cameras of the Magic Leap headset. The goal was to estimate the user's heart rate using the Eulerian Video Magnification algorithm. The project involved understanding the headset's capabilities, implementing the algorithm, and developing a software module to process the eye-tracking camera data. Through testing and refinement, the project successfully estimated heart rate by amplifying color variations caused by blood flow. EyeRate showcased the potential of eye-tracking technology for non-invasive health monitoring.

Research in the Field of Animation Automation

I conducted this research to assess the viability of an MVP designed to streamline the creation process of children's animations. The core idea involved utilizing multiple images of a humanoid animation character as input, with the ultimate goal of producing a well-executed animation featuring said character. Two primary and distinct approaches were explored to achieve this objective.

The first approach centered around harnessing 2D image animation networks. A comprehensive exploration of methodologies falling under this category was conducted to gauge their effectiveness in reducing the time required for animation creation.

Concurrently, the second approach delved into the realm of 3D avatar creation. Notably, the market boasts a variety of sophisticated software capable of infusing lifelike motion into 3D figures. As such, the principal challenge was redirected towards generating 3D avatars from the provided multi-view and multi-pose images of the animated character. This exploration involved techniques adept at seamlessly transforming static images into dynamic, fully-fledged 3D animations, capitalizing on existing motion-enabling software resources.

Education

2019 - 2022

Master's Degree in Electrical Engineering

Bar-Ilan University - Ramat Gan, Israel

2010 - 2014

Bachelor's Degree in Electrical Engineering

Tel Aviv University - Tel Aviv, Israel

Skills

Libraries/APIs

PyTorch, Pandas, NumPy, XGBoost, PyTorch Lightning, Scikit-learn, Matplotlib, OpenCV, TensorFlow, Keras, FFmpeg

Tools

Git, GitHub, Plotly, Algorithm Design, PyCharm, Jira, Pytest, Cron, Open Neural Network Exchange (ONNX), Microsoft Excel, You Only Look Once (YOLO), Bitbucket, Vim Text Editor, Slack, Terraform, Amazon SageMaker, GIS, JMP, PyPI, CAD, Apache Airflow

Languages

Python, SQL, C++, C

Platforms

Ubuntu Linux, Amazon EC2, Jupyter Notebook, Visual Studio Code (VS Code), ClearML, Amazon Web Services (AWS), Linux, AWS Lambda, MacOS, Docker, NVIDIA CUDA

Frameworks

Streamlit, MediaPipe, OpenCL, Realtime

Paradigms

Functional Programming, Unit Testing, Object-oriented Programming (OOP), Automation, Test-driven Development (TDD), Distributed Computing, Testing, DevOps

Storage

JSON, Amazon S3 (AWS S3), Databases, Data Pipelines

Industry Expertise

Healthcare

Other

Image Processing, Deep Learning, Machine Learning, Software, Computer Vision, Data Science, Convolutional Neural Networks (CNNs), Artificial Intelligence (AI), Classification Algorithms, Artificial Neural Networks (ANN), Neural Networks, CSV, Data Analysis, Data Visualization, Monte Carlo Simulations, Random Forests, Supervised Machine Learning, Data Scientist, Startups, Object Detection, CSV File Processing, Machine Vision, Minimum Viable Product (MVP), Model Development, AI Model Training, Machine Learning Algorithms, Neural Network Pruning, Developing AI Models locally, Data Classification, Image Segmentation, Digital Signal Processing, Probability Theory, Research, Computer Vision Algorithms, Calibration, Medical Imaging, PIP, Algorithms, Analytics, Depth Sensors, Software Architecture, Image Analysis, 3D Pose Estimation, APIs, Decision Trees, Statistics, Jupiter, Fine-tuning, AI Programming, Codebase Development, Image Recognition, JupyterLab, 2D, GPU Computing, Graphics Processing Unit (GPU), 2D Modeling, Machine Learning Operations (MLOps), Deep Neural Networks (DNNs), Facial Recognition, Data Loading, Machine Learning Automation, Videos, Data Extraction, Visualization, Health, Data Analytics, Statistical Analysis, XLSX File Processing, Data Cleansing, Data Structures, Data Modeling, Regression Modeling, Large Data Sets, Parallelization, Feature Engineering, Cloud, Pattern Recognition, Writing & Editing, Content Writing, Numba, Python Performance, API Integration, Transformers, Technical Leadership, Data Scraping, Physics, Optical Systems, AI Modeling, AI Research, CSV Export, ONNX Runtime, Infrastructure, Data Inference, AI Consulting, Virtual Reality (VR), AI Integration, Augmented Reality (AR), Quantization, Hugging Face, model quantization, Recurrent Neural Networks (RNNs), Pose Estimation, Data Processing, Optimization, Linear Algebra, Calculus, Applied Mathematics, Hardware, Statistical Methods, Sensor Fusion, Sensor Data, Optical Sensors, Simulations, Architecture, Video Processing, Estimations, Google Colaboratory (Colab), Team Leadership, Back-end, Back-end Development, AWS DevOps, 3D, Consulting, Embryology, NN Compression, Integration, Data Reporting, Optimization Algorithms, Generative Adversarial Networks (GANs), Natural Language Processing (NLP), Avatars, Motion AI, Computer Graphics, Stock Trading, Data Engineering, Digital Elevation Models, Scraping, Chief AI Officer, Production, Electrical Engineering, Llama, Web Scraping, LSTM Networks, FastAPI

Collaboration That Works

How to Work with Toptal

Toptal matches you directly with global industry experts from our network in hours—not weeks or months.

1

Share your needs

Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2

Choose your talent

Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3

Start your risk-free talent trial

Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.

Top talent is in high demand.

Start hiring