Grigor Keropyan, Developer in Munich, Germany
Grigor is available for hire
Hire Grigor

Grigor Keropyan

Verified Expert  in Engineering

Bio

Grigor is an experienced data scientist with a demonstrated work history in the industry and a strong math background, having won several medals in international math olympiads. He also completed two master's degrees with honors and has outstanding problem-solving skills that make him excel in statistics, probability theory, and machine learning. Grigor enjoys being involved in diverse projects and has worked for small teams and large companies like Huawei.

Portfolio

Huawei Technologies Co.
PyTorch, Computer Vision, Machine Learning, Deep Learning...
Technical University of Munich
Python, Machine Learning, Deep Learning, Artificial Intelligence (AI)...
SoftConstruct
Artificial Intelligence (AI), Data Science, Deep Learning, Machine Learning...

Experience

  • Machine Learning - 5 years
  • NumPy - 5 years
  • Data Science - 5 years
  • Pandas - 5 years
  • Python - 5 years
  • Statistics - 4 years
  • Deep Learning - 4 years
  • PyTorch - 3 years

Availability

Part-time

Preferred Environment

Linux, MacOS, PyCharm, GitLab, Slack

The most amazing...

...AI project I've developed is an agent that plays Texas hold 'em poker at a human level in real time.

Work Experience

Computer Vision and Machine Learning Research Intern — Cloud Rendering

2022 - 2022
Huawei Technologies Co.
  • Implemented a real-time denoising algorithm, significantly reducing the time of Monte Carlo rendering.
  • Researched and developed a fully convolutional recurrent autoencoder that removes noise from on-surface caches while keeping temporal information stable.
  • Developed a deep learning Monte Carlo model that removes significant noise from images.
Technologies: PyTorch, Computer Vision, Machine Learning, Deep Learning, Artificial Intelligence (AI), GitLab, Predictive Modeling, Data Visualization, Statistics, Python, Data Engineering, NumPy, C++, Data Science, Jupyter, Linux, Research, GitLab CI/CD, Matplotlib, Scikit-learn, Predictive Learning, ETL, Big Data, Data Analytics, Python 3, Data Analysis, Git

Teaching Assistant

2021 - 2022
Technical University of Munich
  • Prepared and corrected homework assignments for the machine learning course and the machine learning for graphs and sequential data classes.
  • Systematically answered students' questions using Piazza, an online query platform.
  • Corrected final and retake exams for the courses I assisted.
Technologies: Python, Machine Learning, Deep Learning, Artificial Intelligence (AI), Data Science, PyTorch, NumPy, Pandas, Neural Networks, Data Analytics, Data Analysis, Statistics, Jupyter, Matplotlib, Scikit-learn, Linear Regression, ETL, Python 3

Machine Learning Engineer

2019 - 2021
SoftConstruct
  • Developed a counterfactual regret minimization algorithm for Texas hold 'em poker, using reinforcement learning to make the agent play at a human level in real time.
  • Researched and built lossy and lossless abstractions for Texas hold 'em poker.
  • Created a sentence clustering algorithm based on meaning to contribute to the Hoory app's development.
Technologies: Artificial Intelligence (AI), Data Science, Deep Learning, Machine Learning, Python, NumPy, Pandas, GitLab, SQL, Predictive Modeling, PySpark, Data Visualization, R, Statistics, PyTorch, Data Engineering, TensorFlow, C++, Jupyter, PostgreSQL, Linux, Slack, Research, GitLab CI/CD, Matplotlib, Scikit-learn, Predictive Learning, Mathematics, Linear Regression, MongoDB, ETL, BigQuery, Big Data, Spark, Scala, HDFS, Hadoop, Data Analytics, Python 3, Data Analysis, Git

Teaching Associate

2019 - 2020
American University of Armenia
  • Designed and corrected homework assignments for the course on numerical methods.
  • Corrected homework assignments for the ordinary differential equations course.
  • Conducted problem-solving and QA sessions for both courses I assisted.
  • Reviewed mid-term and final exam papers for both courses I assisted.
Technologies: Numerical Methods, R, Python, NumPy, Jupyter, Matplotlib, Mathematics, Python 3, Data Analysis, Data Analytics

Researcher and Software Developer

2016 - 2019
Isp Ras
  • Researched and developed a multiplatform use-after-free and double-free detection framework for binary files based on control flow and data flow graphs.
  • Built a scalable framework for accurate binary code comparison.
  • Created a source code clone benchmark to collect clones.
Technologies: Java, Algorithms, Data Structures, C++, Static Analysis, Machine Learning, Deep Learning, Artificial Intelligence (AI), SQL, Data Science, Statistics, Python, PyTorch, Data Engineering, NumPy, Jupyter, PostgreSQL, Linux, Slack, Research, GitLab CI/CD, Matplotlib, Scikit-learn, Unit Testing, Linear Regression, MongoDB, ETL, BigQuery, Big Data, Spark, Scala, HDFS, Hadoop, Data Analytics, Data Visualization, Python 3, Data Analysis, Git

Experience

Deep Learning Monte Carlo Denoising Model

A fully convolutional recurrent autoencoder that removes noise from on-surface caches while keeping temporal information stable. This algorithm significantly reduces the time of Monte Carlo rendering.

Texas Hold 'em Poker Agent

An AI agent that plays Texas hold 'em poker at a human level in real time. It is currently designed for two-player games, but with more computational power, the algorithm can be trained to play up to six-player matches.

Multimodal Emotion Recognition Model

A deep learning emotion recognition model that predicts emotions and audiovisual signals based on the expressions of the face and voice presented. The project was built from scratch by collecting various datasets and combining them in a network that runs in real time.

Post-nonlinear Causal Models Project

A causal order estimation research project in post-nonlinear causal models for Gaussian noise. The estimation method has been developed and proved consistent in theory, and the proposed practical algorithms are computationally fast and precise.

Binary Code Analysis Tool

https://www.ispras.ru/en/technologies/binside/
A static program analysis platform that finds defects in binary code. This tool can analyze executables and libraries for x86-64, ARM, and MIPS architectures.

The platform detects the following CWE types:
• CWE-121: Stack-based buffer overflow
• CWE-122: Heap-based buffer overflow
• CWE-134: Use of the externally-controlled format string
• CWE-415: Double free
• CWE-416: Use after free
• CWE-77: Command injection

Education

2020 - 2022

Master of Science Degree in Mathematics in Data Science

Technical University of Munich (TUM) - Munich, Germany

2019 - 2021

Master's Degree in Data Science

Yerevan State University (YSU) - Yerevan, Armenia

2015 - 2019

Bachelor's Degree in Informatics and Applied Mathematics

Yerevan State University (YSU) - Yerevan, Armenia

Certifications

JUNE 2019 - PRESENT

Structuring Machine Learning Projects

Coursera

JUNE 2019 - PRESENT

Sequence Models

Coursera

JUNE 2019 - PRESENT

Neural Networks and Deep Learning

Coursera

JUNE 2019 - PRESENT

Improving Deep Neural Networks: Hyperparameter Tuning, Regularization, and Optimization

Coursera

JUNE 2019 - PRESENT

Deep Learning

Coursera

JUNE 2019 - PRESENT

Convolutional Neural Networks

Coursera

Skills

Libraries/APIs

NumPy, Pandas, Matplotlib, Scikit-learn, PyTorch, TensorFlow, PySpark

Tools

GitLab, Slack, GitLab CI/CD, Jupyter, BigQuery, Git, PyCharm

Languages

Python, Python 3, R, SQL, Scala, C++, Java

Paradigms

ETL, Unit Testing

Frameworks

Spark, Hadoop, RStudio Shiny

Platforms

Linux, MacOS, Docker

Storage

PostgreSQL, MongoDB, HDFS

Other

Statistics, Machine Learning, Data Science, Research, Data Analytics, Data Analysis, Data Visualization, Mathematics, Linear Regression, Data Engineering, Deep Learning, Data Structures, Algorithms, Computer Vision, Artificial Intelligence (AI), Static Analysis, Statistical Methods, Statistical Modeling, Big Data, Recurrent Neural Networks (RNNs), Reinforcement Learning, Numerical Methods, Neural Networks, Forecasting, Predictive Modeling, Time Series Analysis, Predictive Learning

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