Led development of an efficient method for placing advertising inserts on various TV advertising platforms within a given budget to maximize audience coverage for Starlink company. The method was based on optimization techniques using genetic algorithms and significantly improved audience coverage, while also operating dozens of times faster than the previous method used by the company.Oversaw the development of algorithmic bots for automated trading on stock markets through broker APIs and the QUIK terminal, using Python language. Worked with Moscow Exchange interfaces using FIX/FAST protocols. Conducted analysis of trading strategies.

Built a medical model development and research ecosystem from scratch, assembling a team of researchers and personally developing key AI solutions. Overcame challenges related to models in production, data management, data labeling, and teaching the labeling process.Recognized the complexities and contradictions within the field of dermatology, what helped me obtain better models. Supervised data annotation efforts and collaborated closely with medical professionals, participating in project pitches to investors.Established collaborations with industry leaders such as GSK, Sanofi, First Moscow State Medical University, and others, contributing to the overall success of the company.Developed AI models for CheckDerm, an automatic system for diagnosing certain dermatological diseases for GlaxoSmithKline (GSK). Created a model for detecting nail diseases for LinkerVerse, a joint Korean company with Scanderm and Seokyeong University. The resulting model was successfully integrated into the HealthyVerse gadget, which won the CES 2023 Innovation Award. Worked on developing models for evaluating dermatological and cosmetic skin conditions for projects such as BeautyScan (Magnit Cosmetics) and FaceScan Pro.My primary research interests here include training on small datasets, data augmentation, anomaly detection, few-shot learning, zero-shot learning, and domain adaptation.

Computer graphics department.Managed a team of four researchers, refining previous ideas and developing a progressive online fine-tuning method for monocular depth prediction models, incorporating optical flow prediction. The method enabled existing models to adapt online in sensors-free manner to previously unseen scenarios where they were ineffective. This adaptation resulted in an average improvement of 30% in the accuracy of the tested models.Conducted research into Long-Term Visual Localization, developing methods for obtaining disentangled representations of photos taken at different times of day, year, etc. Worked on keypoint estimation and matching, and devised a semantic metric scale estimation approach for monocular depth prediction.In my recent work, I have been researching the applications of AI in physical simulation and rendering. Developed a method for interpolating the dynamics of clothing and fabric based on one of the state-of-the-art super-resolution techniques and gradient optimization. The method was several times faster than the GPU version of direct physical simulation.Currently working on research into end-to-end physical simulation using deep learning models.

AR/VR department. Conducted research in the field of Novel View Synthesis for the task of digital twinning of rooms, products, and displays. Analyzed existing scientific literature and developed a novel model for generating new views based on only one RGB image.The model achieved comparable perceptual metric scores and human observation scores to existing state-of-the-art solutions. However, it was designed with a simpler and lighter architecture, resulting in twice as fast inference. The model utilized a CNN CVAE architecture along with a small super-resolution (refinement) model. To enhance performance, custom losses were incorporated from depth estimation experience, including a special perceptual loss.In addition, conducted research into disentanglement issues using different GAN models. By leveraging the GAN's discriminator on the model's latent space, a good approach was found to enhance the final generalization.

AR/VR department. Conducted research in the field of computer vision and deep learning, specifically in unsupervised monocular depth prediction and ego motion estimation. Improved the state-of-the-art method's depth sharpness and convergence time by introducing novel regularization techniques in the model training procedure. The model converged to the same desired result 1.5 times faster on the KITTI dataset, demonstrating the effectiveness of the introduced improvements.

Deliver lectures and seminars on Deep Learning and Mathematical Logic for math students, as well as seminars on Discrete Mathematics for students from other majors.Developed and maintain a personal course on quick introduction to deep learning for undergraduate students, which allows students to choose their own level of immersion and provides relevant theoretical, empirical, and practical knowledge based on the latest research in deep learning.Organized an applied mathematics research circle for students with more than 20 members, focusing on building and supporting a professional researcher community, providing training in current knowledge and research culture, and acting as a platform for finding internships, job opportunities, and commercial projects.Set up the IT infrastructure for the research circle, including an authentication system, messenger, distributed computing system, and other useful services for students and researchers.

Successfully expanded staff and implemented automation initiatives to improve business processes. Led DevOps efforts to enhance system performance and streamline operations. Participated in key negotiations with multiple companies, demonstrating communication and collaboration skills. Directed the development of a virtual mobile operator for security IoT based on MTT telecommunications company, including the creation of a sophisticated billing system and intuitive control panels.

Took charge of identifying and rectifying shortcomings in an outsourced IT provider's performance, resulting in improved efficiency and higher quality deliverables. Developed an entire user protection system, including a mobile application with an interactive SOS button for calling security services, a mobile app for security personnel, and a web-based operator panel. Build a talented development team. Transformed operations by migrating infrastructure to the cloud and implementing cutting-edge Kubernetes technology.

Directed the day-to-day activities of an outsourced IT provider, working closely with internal stakeholders to meet business objectives and exceed expectations. Contributed to the development of a service for calling emergency responders from a mobile device, which included real-time transmission of the user's location and establishment of reliable communication between the user and responders.

Developed various web-sites, small web applications etc