Maksim Kochanov Email & Phone Number

Paris, Fr

• Trained a Transformer-based recommender model on tens of millions of user-track interactions, achieving significant improvements in AB-tests.• Enhanced the second-level recommender model of Gradient Boosting by creating and integrating hundreds of new features, resulting in an improvement in the quality of recommendations by 40% (nDCG@10, mAP@10, and recall@10).• Automated the regular retraining of the Transformer and Gradient Boosting models on new data using Airflow and Kubernetes.• Built ML models of various complexities and delivered them to production environment.

Managed and developed a project to match objects in videos with the appropriate stickers.• Increased the number of detected objects from 200 to unlimited by transitioning from an Object Detection method to Zero-Shot Learning approach.• Employed Vision-Language Model in PyTorch and classical Computer Vision techniques in OpenCV.• Optimized the model architecture by isolating Vision Transformer weights, reducing RAM usage by 20%.• Developed a CPU+IO-bound API service using FastAPI and asyncio in Python, complete with Docker containerization and CI/CD pipelines on GitLab. It’s deployed on thousands of users, handling tens of RPS.• Reduced manual data labeling in the company by over 90% by creating a pipeline that uses the Vision-Language Model and NLP techniques.

Master'S And Engineering Degree, Machine Learning And Computer Vision

Graduating Project: “Hyperspectral Unmixing With Spectral Variability”: • Developed a pipeline in Python and PyTorch to extract.

Bachelor'S Degree, Mathematics And Computer Science, 4.7/5.0

Thesis: “Application of Brain Tumor Segmentation Algorithms to Predict Patient Status”: • Achieved a Dice score of 0.81 with U-Net.