Hyperionide is a startup dedicated to building a cloud-based Python IDE for data scientists.- Developed a hybrid workflow combining the benefits of IDEs with interactive notebooks, facilitating the transition from prototyping to production.- Simplified server operations for streamlined data science resource management.- Integrated AI into the development workflow, including GPT-4, ehnancing the quality of the responses by providing additional context data.-… Show more Hyperionide is a startup dedicated to building a cloud-based Python IDE for data scientists.- Developed a hybrid workflow combining the benefits of IDEs with interactive notebooks, facilitating the transition from prototyping to production.- Simplified server operations for streamlined data science resource management.- Integrated AI into the development workflow, including GPT-4, ehnancing the quality of the responses by providing additional context data.- Designed a framework for frontend development in Python, automatically translating code into JavaScript and WebAssembly. Show less

Aitarget is an advertisement technology company. I was responsible for neural network development and R&D for various adverisement-related tasks.- Developed a service for automatic product description generation based on GPT neural network, utilizing available product data (specifications, images, etc.).- Designed and trained a neural network for speech synthesis (text-to-speech) based on neural transformers, enabling voice generation for audio ads.- Developed neural… Show more Aitarget is an advertisement technology company. I was responsible for neural network development and R&D for various adverisement-related tasks.- Developed a service for automatic product description generation based on GPT neural network, utilizing available product data (specifications, images, etc.).- Designed and trained a neural network for speech synthesis (text-to-speech) based on neural transformers, enabling voice generation for audio ads.- Developed neural networks for object detection in photos; main object identification (e.g., advertised product), and object search by name or textual description; panoptic segmentation. Article about this: https://habr.com/ru/companies/cloud_ru/articles/597017/- Using these neural networks, early versions of Stable Diffusion, and neural networks for image inpainting, I prototyped a product for automatic ad banner editing and reconfiguration for different ad formats. Show less

A startup focusing on indoor positioning. The idea was to combine mobile devices into a sensor network, determining the distance between them based on Bluetooth signal strength, thereby aggregating GPS data to enhance accuracy. After prototype development and initial discussions with potential clients and investors, we decided to discontinue the project.

- Ported several computer vision algorithms from Python to C++ and optimize them for skin analysis devices.- Developed neural networks and computer vision algorithms for skin condition assessment and object detection on skin photographs.- Developed a data annotation system. At the time, HAUT.AI had 10-15 data annotators, so I implemented a web application for annotating our datasets, including object segmentation and classification.- Developed a neural network… Show more - Ported several computer vision algorithms from Python to C++ and optimize them for skin analysis devices.- Developed neural networks and computer vision algorithms for skin condition assessment and object detection on skin photographs.- Developed a data annotation system. At the time, HAUT.AI had 10-15 data annotators, so I implemented a web application for annotating our datasets, including object segmentation and classification.- Developed a neural network architecture based on GAN (WassersteinGAN/ProgressiveGAN + InfoGAN) to edit skin parameters in photographs. This allowed skincare product users to see a forecast of their skin or face appearance after 12 weeks of use.- Developed the first version of a platform for analyzing face/skin photo datasets using our algorithms. Show less

Internship at the Real-time Advertisement Technologies department of Yandex.Worked on the development and optimization of the "banner rotator", a C++ server responsible for selecting and assembling advertising banners displayed in Yandex search and on websites connected to the Yandex advertising network.

2long2read was a startup focused on automatic text summarization.I worked on feature extraction for the main summarization algorithm, and developed an automatic inclusion algorithm using decision trees based on various metrics collected from a webpage's DOM, determining if there was sufficiently long/interesting text to extract main ideas from.