I hold a BEng in Computer Engineering from Wrocław University of Science and Technology.
My research interests are counterfactual explanations, concept-based models, generative models, and LLM interpretability. I am a member of the ML team at Solvro, a student research club at Wrocław University of Science and Technology.
Here you'll find my ML experiments, developer tools, and contributions to Python, Go, Rust, and JavaScript projects.
Original projects · Merged contributions
A selection of my pull requests merged upstream:
| Project | Ecosystem | Summary | Pull Request |
|---|---|---|---|
| PyCQA / isort | Python | Respect extension filters for explicit files with --filter-files |
#2720 |
| urllib3 | Python | Ensure Memray plugin availability before configuring pytest options | #5307 |
| serde_json | Rust | Report invalid UTF-8 at the first bad byte rather than string end | #1345 |
| AnyIO | Python | Skip empty object-stream messages when receiving buffered bytes | #1409 |
| redis-py | Python | Preserve zero-valued GEOSEARCH arguments instead of dropping them |
#4356 |
| Fiber | Go | Let empty optional route parameters yield to later required segments | #4722 |
| narwhals | Python | Avoid trailing separators in concat_str when ignoring nulls |
#3967 |
| pytest-httpserver | Python | Add an opt-in case-insensitive header value matcher | #524 |
| chardet | Python | Respect encoding filters when applying preferred supersets | #393 |
| set-cookie-parser | JavaScript | Ignore invalid Expires attributes to retain valid expiry dates |
#91 |
My DuckDB fix for overwrite=False was submitted in #642 and incorporated by a maintainer in the merged #668.
- UsageTrim: a local CLI and MCP tool that compacts test and build output, with cached text retrievable after recognized secrets are redacted.
- Maintainer Skills Lab: shared maintainer workflows and agent profiles for coding assistants.
- Repro Lens: static reproducibility checks and experiment verification for ML pipelines.
-
Video Waste Dumping Detection (IWDD): video classification with VideoMAE and X-CLIP for the IWDD competition.
-
Sudoku Solver: grid extraction with OpenCV, digit recognition with a PyTorch CNN, and a backtracking solver.
-
GAN Architectures from Scratch: Vanilla GAN, DCGAN, and conditional GAN implementations using PyTorch Lightning and Hydra.
-
YouTube Comment Sentiment Analyzer: a FastAPI and React app for analyzing comment sentiment with Hugging Face models.
-
Adaptive k for Dynamic Classifier Selection: experiments on dataset complexity and neighborhood size in dynamic classifier selection.
-
Local RAG + Reranker: a retrieval CLI using Ollama embeddings, HNSW indexing, and optional cross-encoder reranking.
-
T5 Text Summarizer: T5-small fine-tuning on CNN/DailyMail and XSum, with ROUGE evaluation and a Gradio interface.