Hi, my name is

Roman.

I build cool models!

A passionate Machine Learning Engineer with expertise in deep learning, computer vision, and NLP. I specialize in deploying ML models at scale, optimizing inference performance, and building innovative AI solutions that drive real business impact.

Resume

About Me

I am a Machine Learning Engineer with a passion for building and deploying AI systems that solve real-world problems. I have a strong background in computer science and deep learning, with expertise in computer vision, NLP, and recommender systems.

Currently working at Coinbase, I focus on developing scalable ML solutions that drive business impact. My work spans from optimizing deep learning frameworks to deploying production ML models that save millions in operational costs.

I’m particularly interested in building AI systems that are not only technically excellent but also ethically sound and socially beneficial. I believe in creating technology that genuinely helps people and makes a positive impact on society.

Here are a few technologies I've been working with recently:

Python
PyTorch
TensorRT
Ray
AWS
Kubernetes

Experience

Coinbase
Metaverse Mind Lab
University of Toronto - Vector Institute
Czech Technical University in Prague
Fozzy Group

Machine Learning Engineer - Coinbase

Nov 2023 - Present

Deployed ML model for auto-scaling datastores, saving $2M annually with zero downtime [blog].
Optimized internal deep learning framework using distributed training (Ray), enabling 100x larger datasets, 3x faster training, and up to 25% accuracy gains across four teams [blog]
Implemented TensorRT inference optimization, achieving 12-16x model speedup in serving

Machine Learning Engineer - Metaverse Mind Lab

Sep 2022 - Nov 2023

Simplified cloud infrastructure, reducing deployment time by 50%, securing two enterprise contracts.
Built a real-time question-answering system for VR training, directly supporting $3M+ in new business
Developed core product by integrating LLMs and scene graphs into a novel 3D scene generation system

Machine Learning Research Engineer - University of Toronto - Vector Institute

May 2022 - Aug 2023

Improved bionic leg control accuracy (RMSE) by 20% and achieved 20x faster inference by developing a novel sensor-fusion architecture with optimized serving via TensorRT [paper] [github].
Increased accuracy (BERTScore) of speech decoding from invasive brain implants by 10% by fine-tuning LLMs on neural data [paper]

Deep Learning Engineer - Czech Technical University in Prague

Feb 2022 - Sep 2022

Designed novel multi-modal architecture, improving precipitation nowcasting accuracy by 6%.
Built anomaly detection pipeline for car pricing, reducing pricing error (MAE) by 18%

Machine Learning Engineer - Fozzy Group

Mar 2021 - Mar 2022

Addressed cold-start forecasting by developing a general embedding model for 700K+ retail products.
Built MVP for next-basket recommendation engine, currently serving over 1M users across Ukraine
Improved supply planning accuracy (MAE) by 15% through enhancements in sales forecasting models

Education

Bachelor of Computer Science

National University of 'Kyiv-Mohyla Academy'

Relevant courses: probability theory, calculus, linear algebra, discrete math, advanced graph theory

Selected Publications:

Roman Burakov, Nadyia Shvai, and Bo Wang - Decoding Speech from Electrocorticography with Large Language Models - Bachelor’s Thesis, NaUKMA 2023 [paper]
Oleksii Tsepa*, Roman Burakov*, Brokoslaw Laschowski, and Alex Mihailidis - Continuous Prediction of Leg Kinematics during Walking using Inertial Sensors, Smart Glasses, and Embedded Computing - ICRA 2023 [paper] [github]
Roman Burakov and Tatiana Levitskaya - Automated Processing of Classical Piano Pieces Using a Siamese Neural Network - Science and Industry, PSTU 2018 [paper]