About BNNR
The story, the team, and the technology behind Bulletproof Neural Network Recipe.
What is BNNR?
BNNR (Bulletproof Neural Network Recipe) is an open-source toolkit that automatically finds the best data augmentation strategies for your PyTorch computer vision models.
Instead of manually trying different augmentations and hoping for the best, BNNR runs a systematic search: it trains a baseline model, then iteratively tests candidate augmentations in a branching strategy. Only augmentations that measurably improve performance are kept.
What makes BNNR unique is its combination of novel augmentations (ChurchNoise, ProCAM, DifPresets, and more), XAI explainability (OptiCAM heatmaps showing why the model improves), XAI-driven augmentations (ICD and AICD that use saliency maps to intelligently modify training images), and a real-time dashboard for monitoring every aspect of the search.
BNNR supports both image classification and object detection tasks, with bbox-aware augmentations and detection-specific metrics (mAP@0.5, mAP@[.5:.95]).
Team
Mateusz Walo
Founder & Lead Developer
Architect behind BNNR's core engine, XAI pipeline, and augmentation search algorithm. Passionate about making neural networks more robust and explainable.
Diana Morzhak
Software Developer & QA Engineer
Responsible for feature development, quality assurance, and end-to-end testing — ensuring reliability across classification and detection workflows.
Dominika Zydorczyk
Marketing & Communications Specialist
Drives community outreach, content strategy, and brand presence for BNNR across social channels and developer communities.
Zuzanna Saczuk
Graphic Designer & Brand Lead
Creator of BNNR's visual identity — from the molecular logo and neon branding to UI design and all visual assets.
Tech Stack
MIT License
BNNR is free and open-source software released under the MIT License. Use it freely in personal and commercial projects. Contributions are always welcome.