Professional Work
I currently work at EPR Labs, where I develop software and data pipelines for training, evaluating, and deploying predictive and generative ML models in Python.
Previously, as part of the PRODIS project, I developed the complete ML and data processing infrastructure, including the first phoneme-level GPT model for Polish.
Due to legal and privacy restrictions involving GDPR-protected data, most source code cannot be released publicly. Select components may be open-sourced later.
| Name | Stack | Type | Description |
|---|---|---|---|
| Phoneme-level GPT pipeline | Python, PyTorch, NumPy, Pandas | CLI | Pipeline for training a phoneme-level GPT model to predict surprisal in Polish. Custom IPA tokenizer, parallelized formant extraction, and automatic alignment + stress annotation. |
| Bulk automatic speech recognition | Python, Whisper, FFmpeg | CLI | Pipeline for bulk automatic speech recognition (ASR) using OpenAI Whisper. Also performs stereo-to-mono conversion using FFmpeg. |
| Dataset preparation | Python, Pandas | CLI | Data processing scripts for downloading, cleaning, and phonemizing OpenSubtitles, OSCAR, and Wikipedia corpora into IPA. |
| Survey cleaner | Python, Pandas | CLI | CI-based tool for cleaning and standardizing Microsoft Forms exports with Polish-to-English translation and data validation. |
| Transcriptions status | Python | CLI | CI-based repository scanner that verifies audio transcription structure and generates verification status reports. |
| Word frequency | Python, Pandas | CLI | Data processing scripts for Polish word frequency statistics and X-SAMPA syllabification with cleaning and alignment. |
prodis-opus19.github.io | HTML, CSS, JavaScript | Website | Main website for the PRODIS project, with experiment subpage for collecting data. |
| Fattura | C++17, SFML2 | Desktop | Simple cross-platform GUI app for editing verification statuses with auto-saving and keyboard navigation. |
Personal Projects
| Name | Stack | Type | Description |
|---|---|---|---|
vroom | C++20, SFML3, ImGui | Game | 2D racing game with arcade drift physics, procedurally generated tracks, and waypoint-based AI. |
yt | Python, Streamlit, SQLite, Docker | Web app | YouTube subscription manager. |
header-warden | C++17 | CLI | CLI tool that identifies and reports missing standard library headers in C++ code. |
aegyo | C++17, SFML3 | Desktop | GUI app for learning Korean Hangul. |
ungpt | C++20, SFML3 | Desktop | GUI app that converts ChatGPT’s smart punctuation and symbols to plain ASCII. |
yt-table | C++17 | CLI | CLI for managing YouTube subscriptions. See the newer yt instead. |
asset-packer | C17 | CLI | CLI tool for embedding assets (e.g., images, sounds, fonts) into C++ headers. |
applefetch | C++17 | CLI | CLI system information tool inspired by neofetch. |