• Led end-to-end development of static emotion lenses, including the Crying lens, which achieved the highest engagement in Cameos and has more than 10B views since release.

Machine Learning Engineer in the Generative content department, developing the main features of Snap+ subscription, generating more than 100M$/month:

• Architected and led the development of a quality evaluation system for AI-generated images. This project required the creation of an automated system to assess the output of text-to-image generative AI models. Architected and led the development of generating non-square AI images to align with Snap's vertical screen orientation. This project was essential for maintaining the high quality of Snap’s visual content.

• This solution improved the fidelity of vertically oriented images produced by in-house models, outperforming the existing square-image generation models by more than 30%.

• These two projects not only contributed to the enhancement of Snap’s generative AI capabilities but also led to being named the first author on two patents.

• Developed Re-ID algorithm for football players for WorkNest.

• Developed Event-detection algorithm on airport video for Assaia. Improve speed performance from 1 day training for 1 hour. Also, improve accuracy up to 95%

• Air companies recognition for Assaia.

• Luggage recognition for Assaia.

• Push back detection algorithm for Assaia.

Designed and led end-to-end SafeSearch video classifier improvements. Enhanced by 15% Racy Videos Detection and child abuse video classifier, improved recall by 30% and precision by 17%.

Built an Actor Expansion system to prevent abuse across Google accounts:

• Built and monitored ML pipelines to continuously model training to connect accounts owned by the same actor with a final accuracy of more than 70%, which allowed integration of this system into existing systems.

• Developed end-to-end ETL pipelines to feed data into ML models, that were more than 50 times faster and more reliable than the initial solution.

Led launch of Cloud Baremetal Product.

• Became Focus Area Tech Lead for Monitoring and Health Definition, contributing to improving and optimizing monitoring.

• Decreased rollout time by optimizing risk assessment, which led to sped-up releases by 10%.

• Decreased human delays impact in Orion rollouts via process automatization, which led to sped-up to up to 60% of rollout time.

Led launch of Cloud Baremetal Product.

• Became Focus Area Tech Lead for Monitoring and Health Definition, contributing to improving and optimizing monitoring.

• Decreased rollout time by optimizing risk assessment, which led to sped-up releases by 10%.

• Decreased human delays impact in Orion rollouts via process automatization, which led to sped-up to up to 60% of rollout time.

MEGOGO is the largest video on demand and TV service in the Eastern Europe. Every month we serve more than 20 million of users.

Main responsibilities:

• Create Scene recognition System in video;

• Create Face recognition System in video.

Main technologies: Python 3.6(with ML libraries), OpenCV 3, Tensorflow, Keras.