Avito

Avito is one of the most popular classifieds sites and has an internal platformfor automation of development processes. As a member of the Architecture department's CI/CD team, I developed aninternal CI/CD platform for over 4000 microservices and 1000 engineers. Inaddition to working on the implementation, I was responsible for gatheringfeedback and requirements from business and platform users. I also helpedtransition to new platform features, analyzed customer needs, and adapted theplatform to meet the demand. Finally, I provided engineers with training on theusage of the platform.

MetaMap

MetaMap is an international fintech startup that provides identity verificationservices. The product is a platform for identity verification of users using their biometricand document information. As an ML Infra engineer, I was responsible for thedocument reading part of the platform. My responsibilities included: - Holding tech ownership of the ML platform and migrating it from a monolithicto a micro-services architecture. I designed the architecture, planned themigration, and coordinated tasks with the DevOps department. - Implementing development processes, including build, test, and deploymentpolicies. - Creating a generalized ML toolkit for extracting information from documentimages. This allowed for faster configuration of new ML pipelines for new customers, without the need for engineers. The time frame of this process wasreduced from 6 to 1 week. - Developing an end-to-end pipeline for entity extraction from document scansusing the implemented toolkit on a micro-services environment.

Assaia – The Apron AI

Assaia provides an Artificial Intelligence solution that monitors aircraftturnarounds in real time. I built and led a team of 5 engineers who worked on an internal MachineLearning platform. I established all the necessary processes, such as planning,demos, sprints, roadmapping, code reviews, and performance reviews.My responsibilities included designing the architecture, ensuring codequality, planning, and communicating with business and users. I also activelyparticipated in the development, taking on the combined roles of technical leadand project manager/product owner. The platform consists of several parts: - Tools for collecting, preprocessing, and storing datasets. - A predefined set of tunable deep learning architectures that support bothTensorflow and Pytorch. - A machine learning model registry. - A GPU cluster management system that works on both cloud (supports GCPand AWS) and on-premise. - A reports generation system for analysts and business stakeholders. - A GUI for researchers and analysts. The platform was used for both experiment tracking and training productionmodels. It allowed up to 10 deep learning models to be trained simultaneouslyand to be instantly deployed to production once the reports were approved.