Олег Брагин

ML Engineer

Supervisely OÜ

Computer Vision platform · Tallinn, Estonia · remote

Built and released Live Training and BoT-SORT/ReID features for the Supervisely SDK, and developed dataset analysis and production auto-labeling systems.

• Designed and implemented the Live Training framework—a system for simultaneous data labeling and real-time model training; solved catastrophic forgetting by switching to a transformer architecture and adding custom training-loop stabilization (released in SDK v6.73.506). • Integrated the BoT-SORT tracking algorithm into the Python SDK, including a ReID module to improve object re-identification after occlusions (released in SDK v6.73.419). • Configured and validated a multi-stage auto-labeling pipeline with tracking for production tasks.

Middle ML Engineer

Kaspi.kz

Computer Vision team · largest fintech company in Kazakhstan (NASDAQ: KSPI)

Developed and fine-tuned computer vision models and deployed inference services to production; worked with business stakeholders on metrics and solutions.

• Developed and fine-tuned classification and detection models: DINOv2, YOLOv8, MobileNet v3, EfficientNet v2; image generation with Stable Diffusion and Flux; experiment tracking in MLFlow. • Developed and deployed inference services to production (FastAPI, Docker). • Worked directly with business stakeholders: defining metrics, presenting solutions.

ML Engineer (part-time)

Mathshub EdTech

Developed and deployed a student churn prediction product; course completion reached 80%, compared with an approximately 20% industry average.

• Developed an ML pipeline for student churn prediction, deployed as a live product; course completion rate increased to 80% against an industry average of about 20%. • Built an XGBoost + Random Forest ensemble on behavioral data. • Used LSTM and YandexGPT API for NLP analysis of student communications (engagement, signs of stress). • Conducted exploratory data analysis and worked with Pandas.

ML Engineer

Freelance (commercial project)

Built an end-to-end 3D reconstruction pipeline combining COLMAP/GLOMAP, SAM3, and 2D Gaussian Splatting.

• Built an end-to-end pipeline for 3D reconstruction of objects from video or multi-view photos into a finished textured 3D model. • Combined open-source tools into a working chain: COLMAP/GLOMAP for geometry reconstruction from photos, SAM3 for object segmentation, and 2D Gaussian Splatting for rendering.

Professional training program, ML Engineer

Mathshub

Courses

Deep Learning School, MIPT (Moscow Institute of Physics and Technology)

Specialist degree (5-year program), Economist (Economic Security)

Siberian Federal University