Zryw
Workout tracker: log a set in two taps, follow a training plan and get feedback from an AI coach that reads your own log. Works offline. iPhone and Apple Watch first; Android, Wear OS and a web app are in progress.
Workout tracker: log a set in two taps, follow a training plan and get feedback from an AI coach that reads your own log. Works offline. iPhone and Apple Watch first; Android, Wear OS and a web app are in progress.
AI CV builder. Paste a job offer: rezme tailors your CV to it, drafts a cover letter and keeps the result readable for applicant tracking systems (ATS).
Dictation: press a hotkey, speak, and the text is pasted into any app. Runs on your own AI key or offline Whisper models, no account. Desktop 1.5.0 for Windows, macOS and Linux; iOS in TestFlight, Android coming.
My personal AI assistant: a .NET hub on a VPS, an Electron desktop client with a holographic HUD across monitors, and an iOS app. It also runs Sentinel, a security scanner for my servers and repos, and Studio, a video editor and demo recorder.
Desktop app for analysing fluorescence spectra of blood serum (EEM / PARAFAC). Built for Zuzanna Buryan, who holds the copyright; I develop it.
Blish HUD module for Guild Wars 2. Reads your account through the official API and plans the next steps toward a goal such as a legendary weapon: requirements, materials and trading post prices.
Vertical (9:16) video editor for iPhone, iPad and Mac with on-device AI, built only on Apple frameworks.
Senior Software Developer based in Silesia, Poland. About five years of commercial work, the last three in AI/ML.
Day job: enterprise AI assistants and RAG systems on Azure. I work in .NET and Python with Semantic Kernel and Azure AI Search, and I wire in OpenTelemetry tracing so every answer can be traced back to the retrieval and model calls behind it.
After hours: I build and ship my own products (Zryw, Whisperio, rezme, Jarvis) with an AI agent setup that does much of the implementation while I review and test.
Every change goes through a build, tests and a run of the real flow before it counts as done.
LLM features get evaluations and tracing. If a plain rule does the job, it stays a plain rule.
Apps go to real stores and servers with real users, then get better from feedback.