Daria Perina

CV/ML Engineer with expertise in Adversarial Machine Learning and AI Security. Specializing in synthetic content detection, video analytics, and the security of AI systems (Adversarial ML). Experienced in deepfake generation research, analysis of open-source repositories, and evaluating model robustness against adversarial attacks.

Designed and implemented identity-swapping generators (FaceSwap / Diffusion Models) for both research and commercial applications. Build end-to-end ML pipelines—from data collection and preprocessing to model training and production evaluation. Master's degree in Computer Science / Software Engineering with a GPA of 4.85/5.0.

Core technology stack: Python, PyTorch, PyTorch Lightning, OpenCV, MediaPipe, Docker, Linux, Git. Currently seeking a project-based or full-time position in an ML team. Interested in challenging R&D tasks at the intersection of Computer Vision, Generative AI, AI Security, and industrial video analytics.

CV / ML Engineer (R&D) at IDscan.net Inc. (USA) (2025-09 – Present)

Development and research of a commercial system for detecting generative artifacts and deepfakes (Deepfake Detection).

• Built a large-scale training dataset containing more than 40,000 videos, successfully integrating several industry benchmarks and internal company datasets.
• Developed identity-swapping generators and created synthetic datasets based on them.
• Designed an automated video preprocessing pipeline including face detection, five-point alignment, streaming artifact filtering, and frame extraction.
• Designed and evaluated two data analysis approaches: spatial (frame-level) and temporal (sequence-level).
• Conducted a series of experiments on training and optimizing state-of-the-art architectures, including lightweight CNNs and Vision Transformers.

CV / ML Engineer at Smart City / Video Analytics (2024-08 – 2025-09)

• Developed Python scripts using OpenCV for automated video cleaning, including the removal of corrupted, black, and blurry frames.
• Implemented a semi-automatic annotation pipeline: trained an object detection model and annotated the dataset in batches, creating a dataset containing more than 20,000 images.

Master's Degree in Software Engineering (Honors) – Chelyabinsk State University IIT (2026)