Senior Computer Vision and Deep Learning Engineer

Svitla Systems Inc. is looking for a Senior Computer Vision and Deep Learning Engineer for a full-time position (40 hours per week) in Ukraine. Our client is an American mobile computing company.

The ideal candidate combines strong programming skills, hands-on experience training and deploying deep learning models, and a systematic approach to failure analysis. You will work with scene text and document understanding, practical multi-view geometry, and rigorous benchmarking, as part of a team collaborating on shared codebases and remote projects.

Requirements

• Strong programming skills in Python and key libraries such as NumPy and OpenCV.
• Hands-on experience training and deploying deep learning models in PyTorch.
• Solid background in scene text and document understanding: text detection (segmentation- or regression-based), text recognition (CTC-, attention-, or transformer-based), sequence decoding and confidence calibration, grammar- or lexicon-constrained decoding, and reading-order and layout grouping for key information extraction.
• Practical multi-view geometry experience: local feature detection, description, and matching; homography estimation; robust estimation such as RANSAC and its variants; inlier analysis; and perspective warping.
• Ability to debug chained coordinate-space transforms and polygon geometry.
• Systematic failure analysis skills: building an error taxonomy over a pipeline, turning anecdotal failures into reproducible test cases and quantified failure buckets, and running controlled ablations that isolate the responsible stage.
• Experience with dataset construction and curation, augmentation, synthetic and degraded data generation, domain adaptation, fine-tuning, and hyperparameter sweeps.
• Discipline to benchmark rigorously before and after every change against tracked baselines.
• Proficiency with Hydra for configuration management and Weights & Biases for experiment tracking.
• Experience working on remote GPU servers and clusters and in containerized environments.
• Experience with Git and GitHub for version control.
• A habit of testing numerical and geometric code.
• Proven ability to collaborate effectively in a team, particularly on shared codebases and remote projects.
• Clear technical writing, including analysis reports that support decision-making.

Responsibilities

• Own and improve the OCR pipeline end to end.
• Diagnose failures on real customer images and identify root causes.
• Develop and optimize text detection, recognition, decoding, and layout components.
• Work with image geometry, feature matching, homographies, RANSAC, and perspective transforms.
• Build and curate datasets, run augmentations and fine-tuning, and conduct controlled experiments.
• Benchmark changes against tracked baselines and document measurable improvements.
• Use PyTorch, OpenCV, NumPy, Hydra, Weights & Biases, GPUs, containers, and Git.
• Collaborate with the team on shared codebases and provide clear technical analysis.