Skip to content
DSML.KZ
AT

Anton Taryanik

@Toxele

Machine Learning Engineer, Computer Vision and Multimodal Retrieval

Machine learning engineer with 2+ years of experience. I like solve tasks from deep learning such as CV or RecSys

Career

Experience: 2 years

Experience

Laboratory Research Assistant

ITMO University Laboratory

Research laboratory at ITMO University.

Sep 2025 — Present

Developed synthetic Cyrillic data tools and a lightweight recognition model with 99.6% digit accuracy.

• Developed a synthetic EMNIST-like dataset generation library for Cyrillic characters using OpenCV and PyTorch. • Achieved 99.6% digit-recognition accuracy with a lightweight approximately 4 KB model. • Conduct experiments on Cyrillic character recognition, synthetic data generation and model efficiency. • Preparing results for a scientific publication.

Backend Developer

Anizo

AI-based video and audio monitoring startup

Mar 2026 — Jun 2026

Built backend features, aggregated APIs, access controls, and tests for an AI monitoring platform.

• Developed backend functionality for an AI-powered video and audio monitoring platform for early detection of abnormalities in animals and fish. • Designed aggregated API endpoints around frontend use cases, reducing client-side data orchestration. • Implemented role-based access control and improved access separation across backend functionality. • Refactored parts of the existing codebase and developed tests for API endpoints and local database interactions.

Education

Bachelor, Engineering of Artificial Intelligence

ITMO University

Scientific and Practical Intensive Course, Data Analysis

Sirius University

Selected projects

AI Challenge 2026, Image Manipulation Localization

Researched segmentation models for pixel-level localization of manipulated image regions, achieving a confirmed leaderboard AIC of 0.94208 and validation Dice of approximately 0.917 under strict inference constraints.

AI Challenge, Multimodal Retrieval for CAD Objects

Built image and text retrieval systems for CAD models, achieving Recall@5 of 0.99 on the public leaderboard and 0.94 on the private leaderboard for image-to-image retrieval, and 0.74 for image-text retrieval.

Perspective Distortion Correction for Document Images

Developed a model using synthetic training data and evaluated generalization on the real-world WarpDoc dataset.

Straight Line Detection in High-Noise Screenshots

Developed a model using synthetic pretraining and fine-tuning on real data, achieving mean angular error below 0.5 degrees per image.

Achievements

2nd place, IT Purple Hack, overall result across Sber, Avito and I.g.S cases

IT Purple Hack · 2026

7th place, AI Challenge 2025, Multimodal Retrieval for CAD Objects

AI Challenge · 2025

Skills

PyTorchscikit-learnPySparkOpenCVYOLOimage segmentationmetric learningmultimodal retrievalclassificationsegmentationsynthetic data generationrepresentation learningimage retrievalTransformersFAISSLangChainPythonGoC++Rust