I work in the Proof of Concept team at Chaptr.AI, testing new AI approaches together with publishing houses. Before that I worked on Reedy, and before Chaptr I was on the team that shipped TRURL, Poland's first large-scale generative model.
Outside work I write fiction. Since October 2024 I have been working on a dark fantasy novel, which I am now editing.
Research, open-source, and industry work
Modern prompt engineering, designed the way a human would — clean, minimal, elegant. An open-source skill for Claude and Cursor that encodes best practices for structured outputs with Pydantic models.
A mystery game where you play as a cat in an apartment block. I started it at Vibe Jam 2026 and it is now a full game of about two hours, with a scent-based investigation, seven chapters and two endings. Built with Three.js for desktop and mobile.
PhD thesis and NCN PRELUDIUM 18 grant on bias detection, XAI, and mitigation; skin-lesion case studies, style transfer / targeted augmentation / attribution feedback; IEEE Access survey with M. Grochowski; GEBI; MICCAI 2022 GAN debiasing in dermoscopy.
An AI engine at Chaptr.AI that generates keywords, THEMA codes and SEO descriptions for book catalogs. I worked on it before moving to the Proof of Concept team.
Trained models across computer vision (classification, detection, segmentation), NLP and generative text, and audio; with the VoiceLab team shipped TRURL 7B/13B (incl. 8-bit and academic), vlt5-base-keywords, herbert-base-cased-sentiment, and datasets on Hugging Face.
Curated review of augmentation techniques, libraries, and papers (1.6k+ stars), plus research code for Targeted Data Augmentation (TDA) and Counterfactual Bias Insertion (CBI) on dermoscopy and face data.
Dermoscopy / melanoma bibliography (classical features through NAS, self-supervision, augmentation, bias) plus deep learning for other biomedical imaging—microbleeds, blood smear, erythrocytes—with links added over time.
WiMLDS Trójmiasto AI4Good: PyTorch detection and segmentation (EfficientDet, DETR, Mask R-CNN, Faster R-CNN) on merged litter benchmarks, Waste Management paper, curated waste image datasets list (338+ stars), and co-organized Hack4Environment (waste & environmental literacy with DIH4.AI).
Making the world more accessible for the Deaf community through deep learning-based sign language recognition.
Created the WikiPunct dataset and organized the first Polish punctuation restoration shared task at PolEval 2021.
Generating 64×64 retro-pixel characters using Generative Adversarial Networks.
Community-maintained reference of ML and AI acronyms and abbreviations.
WiMLDS Trójmiasto project — sound-based bird species classification using deep learning on audio spectrograms.
For work, collaboration or speaking, message me on LinkedIn. Outside work I write, read, cook and spend time with my two cats.