I’m Nikita Sokovnin, a machine learning engineer in Prague with 5+ years of experience shipping production computer vision systems and applications. I focus on inference optimization, scalable ML pipelines, and tools that speed up real-world delivery—including DataDreamer, an open-source project for automated dataset generation with foundation models and GenAI.

I’ve worked across research and production: at CIIRC I contributed to robot perception and RL (including the myGym toolkit and work published at ICCAR 2021); at Luxonis I built and deployed YOLO / SAM / VLM-style pipelines with Docker, Kubernetes, and GCP; since late 2025 I’m at EPAM continuing hands-on ML engineering.

Education: Master’s in Computer Vision (CTU FEL) — thesis on open-vocabulary detection with multimodal and generative models; KAIST exchange in AI; BSc with thesis on open-set 3D object detection; graduate ML at Charles University (prg.ai Minor).

Outside of work I stay curious about RL, detection and segmentation, and new AI tooling. I’m a member of Upsilon Pi Epsilon, enjoy hackathons, hiking, skiing, and jogging, and I like computer and board games.

For the full timeline and skills, see CV. For code, GitHub; for publications, use the link on the CV page (Semantic Scholar).