Sep 2023 – Jul 2024 · 10 mos · Ottawa, Ontario, Canada · Remote

• Helped multiple local startups with their data-related problems including:
• Implemented an algorithm from paper to re-project 3D volumetric CBCT data into 2D XRay-like images;
• Tested performance of different open source speech-to-text systems using Docker containers and written comprehensive report with recommendations;
• Consulted a tech startup about how they can solve their business problems using Pose Estimation techniques

Jul 2022 – Sep 2023 · 1 yr 2 mos · Vancouver, British Columbia, Canada

• Worked on state-of-the-art Computer Vision research related to Makeup Transfer and Virtual Try-On
• I have developed an innovative approach for transferring makeup, encompassing lipstick, eyeshadow, and foundation, by leveraging synthetic data for training. The model produces significantly more visually appealing outcomes compared to the leading existing models
• Established backend infrastructure using EC2 and FastAPI to facilitate model serving for demo applications, enabling users to dynamically alter their makeup in real-time during Virtual Try-On experiences.
• Built pipeline in AWS for generating large dataset of synthetic images of human faces with corresponding labels using Blender and Unreal rendering engines. This dataset supported various deep learning initiatives, including training of skin care recommendation models

Aug 2020 – Jul 2022 · 1 yr 11 mos

• Supervising work of a small team (5 FTEs+3 interns) delivering results on multiple projects, including:
• Contextual Monitoring project - created a development platform and custom visualization web tool (using React+d3.js+Flask stack) for investigators to enable research on graph data of interactions between bank‘s corporate clients; lead research on applying graph convolutional neural networks to capture AML customer risk
• Mexico Customer Segmentation - Set up machine learning pipeline and trained customer segmentation model using clustering techniques that is estimated to reduce 20% of false positive transaction monitoring alerts
• Investigator Toolkit website - Massively improved user experience by reducing backend data load time 50x and implementing a lot of new features by extending the Dash framework with custom Javascript code; written Bash/Python scipts to automate ETL process and make it run fail-proof on very challenging infrastructure

Sep 2018 – Aug 2020 · 1 yr 11 mos

• Developed risk models for regulatory parameter estimation in compliance with Bazel II guidelines which were applied to Trade Receivables and Auto Loans Securtization exposures;
• Implemented complex modelling logic using Python OOP features with pandas, numpy, sklearn packages;
• Rewritten bottleneck places of code in C leading to 200x boost in execution time allowing to train more interesting models

Mar 2019 – May 2020 · 1 yr 2 mos · Remote

• Punch-to-Face is a startup aiming to reconstruct MMA fights in VR
• Lead a team of 12 researchers and developers bringing the startup from zero to the launch stage. The demo-video https://youtu.be/l_4FK8nBmEA demonstrates our achievements;
• Created and implemented a novel algorithms for calibration of moving and zooming camera combining classical computer vision techniques from OpenCV library and global optimisation techniques from sciPy library;
• Supervised collection of a dataset and trained a semantic segmentation model to segment fighters, canvas and advertisements using PyTorch;
• You can read more about what I built there in this Medium article https://medium.com/@andriylvtsk/making-an-augmented-reality-product-using-computer-vision-1cfe3b5741ab

Aug 2017 – Jun 2018 · 10 mos · Greater Toronto Area, Canada

• Gave tutorials to a second-year macroeconomics class;
• Coordinated with other TAs to initiate the use of LaTex presentations to improve learning outcomes

Apr 2016 – Sep 2016 · 5 mos · Greater Cambridge Area

• Work included data preparation tasks for big research projects in area of political economics;
• My suggestions helped to shape the direction of research;
• Used R automations to speed up the work and to detect mistakes in data;
• Used advanced Excel functionality to prepare templates for future data entry;
• Coordinated remote work of 3 team members.