# Mateusz Śliwa

> Data and AI engineer who ships mobile apps. MSc in Data Engineering (University of Zielona Góra, defended 2026) and BA in economics from the same university. Based in Zielona Góra, Poland; open to remote or hybrid roles and to relocation; available now. Languages: Polish (native), English (C1, Cambridge C1 Advanced 2025), German (A2). Email: sliwa.mateo@gmail.com. GitHub: https://github.com/MSliwa01.

He builds with coding agents (Claude Code, Codex, Cursor), writes agent instructions (AGENTS.md, CLAUDE.md) and skills, and builds tools that agents can drive. Inside his products he uses Gemini in the EcoSort reading pipeline and Gemini with a Gemma fallback in the EcoSort assistant. wwxd runs as a Claude Code plugin.

Last updated: 2026-10-13. Every project below has its own page with a working web demo and the full write-up; the EcoSort and thesis pages show how their numbers were measured. Add /pl before the path for Polish (for example https://mateuszsliwa.dev/pl/ecosort).

## CV

- [CV in Markdown](https://mateuszsliwa.dev/cv.md): projects, skills, education and languages, in English
- [CV po polsku](https://mateuszsliwa.dev/cv.pl.md): the same in Polish
- [JSON Resume](https://mateuszsliwa.dev/resume.json): the CV in the JSON Resume 1.0 format
- [CV as PDF](https://mateuszsliwa.dev/files/Mateusz_Sliwa_CV_DataAI_EN.pdf): one page, the version to forward (short link: https://mateuszsliwa.dev/cv)

## Projects

- [EcoSort](https://mateuszsliwa.dev/ecosort): iPhone app on the App Store in Poland (since June 2026) that shows waste collection dates for your street, with widgets, reminders and a Live Activity. Its LLM pipeline reads scanned municipal schedules one month at a time: two reads with different prompts, a third read on disagreement, and code checks. On six hard scans (2,190 cells) the best of three single prompts got 24.5% of cells wrong. Over three full runs of the pipeline (6,570 cell results) none was a wrong date: 82.2% were published and 17.8% were held back instead of guessing. The pipeline was tuned on these documents, so this is the in-sample best case; a test on new documents is still to come. The Android app is built but not yet on Google Play. App Store: https://apps.apple.com/pl/app/ecosort/id6785674008 . Website: https://ecosort.pl
- [Keenwit](https://mateuszsliwa.dev/keenwit): brain-training app in development with 43 short games. React Native and Expo on a Fastify and PostgreSQL backend; the server checks every result before XP, streaks or rankings move.
- [TrueSky](https://mateuszsliwa.dev/truesky): weather drawn as a painted landscape (Expo, React Native, Skia), free with a monthly or lifetime Pro purchase through RevenueCat. On the App Store since 9 October 2026: https://apps.apple.com/pl/app/truesky-weather/id6816346955
- [MSc thesis](https://mateuszsliwa.dev/thesis): win probability in Counter-Strike 2 from 16.8 million game states in 187,361 rounds of 9,196 professional maps. Nine models compared with cross-validation grouped by match; gradient boosting came out best. In quarterly walk-forward tests XGBoost reached ROC-AUC 0.81 to 0.82 once it had three quarters of history. The production XGBoost model has a calibration error (ECE) of 0.009.
- [Skin lesion classifier](https://mateuszsliwa.dev/skin): machine learning course project on HAM10000 (PyTorch). A fine-tuned EfficientNet-B0 reached AUC 0.93 and 0.95 recall on malignant lesions; a custom CNN reached AUC 0.755.
- [wwxd](https://mateuszsliwa.dev/wwxd): "What would X do?" Open-source (MIT) Python CLI and Claude Code plugin. It ingests one person's public captions, podcast transcripts and articles into a wiki of their positions, and an agent answers with timestamped quotes. A script checks every quote against the transcript and found 158 of 160 cited quotes word for word (the 2 misses were video titles). On 40 questions a blind judge from another lab (Google) preferred it to plain Claude on 34 of 40. Panel mode answers for several people at once and maps where they agree and split. It does not predict opinions a person hasn't voiced yet. Code: https://github.com/MSliwa01/wwxd
- [storeshots](https://mateuszsliwa.dev/storeshots): open-source CLI that makes App Store screenshots from the running app; every command prints JSON so a coding agent can drive it. Code: https://github.com/MSliwa01/storeshots
- [macsim](https://mateuszsliwa.dev/macsim): open-source tool that turns a Mac on the network into a remote iOS build server and iPhone Simulator for a Linux machine, so coding agents can build, run and test iOS apps from Linux. Code: https://github.com/MSliwa01/macsim

## Contact

- [Email](mailto:sliwa.mateo@gmail.com): sliwa.mateo@gmail.com
- [GitHub](https://github.com/MSliwa01): MSliwa01
- [Website](https://mateuszsliwa.dev/): mateuszsliwa.dev, in English; Polish at https://mateuszsliwa.dev/pl/