At Amazon, I author classification logic that powers internal ML/LLM systems for detecting restricted products across EU marketplaces. I validate model outputs and refine rules to reduce false positives.
In Amazon's Electronics Classification work, I used AutoML and LLM-based systems to improve classification accuracy by 35% and cut manual review time by 50%. I also classified more than 100,000 products while maintaining 99%+ compliance accuracy.
I build open-source AI projects, including Unchained, a single-file agentic framework with tool-calling, memory, RAG, and multi-agent orchestration. Polyglot Swarm is a self-hostable AI language tutor, while my chess coach project pairs Stockfish evaluations with an LLM faithfulness harness.
I hold an MSc in Automatic Control, Cybernetics and Robotics from Gdansk University of Technology, with a specialization in Decision Systems and Robotics. I'm also a co-author of an award-winning IEEE HSI 2024 paper on an AI-powered cleaning robot.
