Harut Margaryan, Developer in Yerevan, Armenia
Harut is available for hire
Hire Harut

Harut Margaryan

Verified Expert  in Engineering

AI and Software Developer

Yerevan, Armenia

Toptal member since April 27, 2022

Expertise
Computer VisionMachine LearningDeep LearningArtificial IntelligenceDebuggingLinuxPythonPyTorchGitDockerMongoDBFlaskTensorFlow
Bio

Harut is a computer vision (CV) engineer specializing in deep learning and solving problems by integrating CV models into real-life applications. He built an unsupervised image classification algorithm used in canteens to classify food on trays and a new method of planning spacecraft flights with reinforcement learning and Monte-Carlo tree search simulations, finding trajectories that would have saved fuel on the Voyager 2 flight. Harut is proficient in Python, Linux, and model deployment.

Portfolio

ArtMind
Python 3, Docker, PyTorch, MongoDB, Shapely, GeoPandas, Unity3D, C...
SmartClick
Python 3, PyTorch, MongoDB, Flask, Scikit-learn, OpenCV, Computer Vision...
HighInt
Computer Vision, Python 3, Keras, TensorFlow, Object Detection...

Experience

  • Computer Vision - 3 years
  • Python 3 - 3 years
  • PyTorch - 3 years
  • Git - 3 years
  • Linux - 3 years
  • Docker - 2 years
  • Flask - 2 years
  • MongoDB - 2 years

Availability

Part-time

Preferred Environment

PyCharm, Linux, Docker, MongoDB

The most amazing...

...tool I've developed is an innovative algorithm to fill an empty apartment with furniture, taking into account user preferences.

Work Experience

Machine Learning Engineer

2020 - PRESENT
ArtMind
  • Developed the AI portion (mainly Computer Vision) of a mobile application to create an apartment design.
  • Trained, deployed, and maintained machine learning models using Python and Docker containerization.
  • Proposed and implemented a new algorithm to fill the room with furniture, taking the user's preferences into account.
  • Proposed and implemented a genetic algorithm to divide furniture (such as wardrobes) into separate sections. A user can buy them in a store and spend less money than if he ordered sections with exact dimensions.
  • Implemented a pathfinding algorithm in the C programming language to make the generated apartment fulfillments passable.
  • Used R-tree data structure for indexing the polygons of rooms, with which I sped up filling algorithm five times.
  • Implemented Monte Carlo tree search algorithm for selecting the best combination of candidates for each room in the apartment.
Technologies: Python 3, Docker, PyTorch, MongoDB, Shapely, GeoPandas, Unity3D, C, Genetic Algorithms, Monte Carlo Simulations, Machine Learning, Object Detection, Python, Redis Cache, Protobuf, Deep Learning, SciPy, Git, Unit Testing, Debugging, Profiling, Software Design Patterns, Discrete Optimization, Web Scraping, Selenium, Beautiful Soup, TensorFlow, Artificial Intelligence (AI), Image Recognition

Computer Vision Engineer

2019 - 2020
SmartClick
  • Developed the Computer Vision portion of an application for recognizing food on trays in canteens and automatically calculating prices without the help of a cashier.
  • Proposed a new image clustering algorithm using self-supervised learning, which works with many classes and a small number of instances for each class.
  • Proposed an approach to training a food classification model that did not require re-training each time new classes were added.
  • Completed deep research in the field of Few-Shot Learning to train a classification model using only 5-10 instances for each class of food.
Technologies: Python 3, PyTorch, MongoDB, Flask, Scikit-learn, OpenCV, Computer Vision, Python, Deep Learning, SciPy, Object Detection, Git, Unit Testing, Software Design Patterns, Few-shot Learning, Clustering, Artificial Intelligence (AI), Image Recognition

Computer Vision Engineer

2019 - 2019
HighInt
  • Trained object detection models for satellite imagery.
  • Configured the Jetson Nano Developer Kit to run object detection models on it.
  • Compared the performance of several trained models on Jetson Nano.
Technologies: Computer Vision, Python 3, Keras, TensorFlow, Object Detection, Artificial Intelligence (AI), Image Recognition

Experience

ArtMind

https://www.youtube.com/watch?time_continue=1&v=8v6Ph7dq1Sk&feature=emb_logo
A mobile application for generating designs for apartments using AI. I developed the AI portion of the project. There were four developers on the team, two front end, one back end, and myself, a machine learning engineer. I created microservices with ML models and provided API for front-end developers to work with the models.

Few-shot Food Recognition

I developed the computer vision portion of the software that recognizes food on a tray in canteens and calculates prices without the help of a cashier. The software includes real-time instance segmentation, few-shot image classification, and object tracking.

Visual Product Search

I implemented the Computer Vision potion of a mobile app to recognize apparel items on a photo and provide a link to look-alike items at an online store. The project was created during the AI hackathon and took first place.

Interplanetary Flights Planning

https://www.youtube.com/watch?v=uogiCQ56Qaw
This project was part of my thesis. During my research, I came up with a new way of planning spacecraft flights with the help of reinforcement learning and Monte-Carlo tree search simulations. I tested my method on the actual flight data of the Voyager 2 spacecraft and was able to find a solution that would save fuel if Voyager 2 followed a different trajectory.

Education

2015 - 2019

Bachelor's Degree in Mathematics and Computer Science

Lomonosov Moscow State University - Yerevan, Armenia

Skills

Libraries/APIs

PyTorch, Shapely, Scikit-learn, SciPy, TensorFlow, Beautiful Soup, OpenCV, Keras, Protobuf

Tools

LaTeX, Git

Languages

Python 3, Python, C#, Lisp, C, Java, SQL

Frameworks

Flask, Selenium, Unity3D

Paradigms

Unit Testing

Platforms

Linux, Docker, Software Design Patterns

Storage

MongoDB, SQLite, Redis Cache

Other

GeoPandas, Genetic Algorithms, Monte Carlo Simulations, Processing.js, Computer Vision, Machine Learning, Object Detection, Deep Learning, Debugging, Profiling, Probability Theory, Mathematics, Few-shot Learning, Clustering, Discrete Optimization, Artificial Intelligence (AI), Image Recognition, Web Scraping, FAISS

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